release: prepare for 4.0.11

* seastar 065a40b34a...748428930a (1):
  > append_challenged_posix_file_impl: allow destructing file with no queued work

Fixes #7285.

.gitmodules

@@ -15,3 +15,6 @@
 [submodule "zstd"]
 	path = zstd
 	url = ../zstd
+[submodule "abseil"]
+	path = abseil
+	url = ../abseil-cpp

SCYLLA-VERSION-GEN

@@ -1,7 +1,7 @@
 #!/bin/sh
 
 PRODUCT=scylla
-VERSION=4.0.3
+VERSION=4.0.11
 
 if test -f version
 then

alternator/conditions.cc

@@ -365,31 +365,35 @@ bool check_compare(const rjson::value* v1, const rjson::value& v2, const Compara
 
 struct cmp_lt {
     template <typename T> bool operator()(const T& lhs, const T& rhs) const { return lhs < rhs; }
+    // We cannot use the normal comparison operators like "<" on the bytes
+    // type, because they treat individual bytes as signed but we need to
+    // compare them as *unsigned*. So we need a specialization for bytes.
+    bool operator()(const bytes& lhs, const bytes& rhs) const { return compare_unsigned(lhs, rhs) < 0; }
     static constexpr const char* diagnostic = "LT operator";
 };
 
 struct cmp_le {
-    // bytes only has <, so we cannot use <=.
-    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return lhs < rhs || lhs == rhs; }
+    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return lhs <= rhs; }
+    bool operator()(const bytes& lhs, const bytes& rhs) const { return compare_unsigned(lhs, rhs) <= 0; }
     static constexpr const char* diagnostic = "LE operator";
 };
 
 struct cmp_ge {
-    // bytes only has <, so we cannot use >=.
-    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return rhs < lhs || lhs == rhs; }
+    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return lhs >= rhs; }
+    bool operator()(const bytes& lhs, const bytes& rhs) const { return compare_unsigned(lhs, rhs) >= 0; }
     static constexpr const char* diagnostic = "GE operator";
 };
 
 struct cmp_gt {
-    // bytes only has <, so we cannot use >.
-    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return rhs < lhs; }
+    template <typename T> bool operator()(const T& lhs, const T& rhs) const { return lhs > rhs; }
+    bool operator()(const bytes& lhs, const bytes& rhs) const { return compare_unsigned(lhs, rhs) > 0; }
     static constexpr const char* diagnostic = "GT operator";
 };
 
 // True if v is between lb and ub, inclusive.  Throws if lb > ub.
 template <typename T>
 bool check_BETWEEN(const T& v, const T& lb, const T& ub) {
-    if (ub < lb) {
+    if (cmp_lt()(ub, lb)) {
         throw api_error("ValidationException",
                         format("BETWEEN operator requires lower_bound <= upper_bound, but {} > {}", lb, ub));
     }

alternator/executor.cc

@@ -565,7 +565,7 @@ static void validate_tags(const std::map<sstring, sstring>& tags) {
 // to races during concurrent updates of the same table. Once Scylla schema updates
 // are fixed, this issue will automatically get fixed as well.
 enum class update_tags_action { add_tags, delete_tags };
-static future<> update_tags(const rjson::value& tags, schema_ptr schema, std::map<sstring, sstring>&& tags_map, update_tags_action action) {
+static future<> update_tags(service::migration_manager& mm, const rjson::value& tags, schema_ptr schema, std::map<sstring, sstring>&& tags_map, update_tags_action action) {
     if (action == update_tags_action::add_tags) {
         for (auto it = tags.Begin(); it != tags.End(); ++it) {
             const rjson::value& key = (*it)["Key"];
@@ -592,24 +592,12 @@ static future<> update_tags(const rjson::value& tags, schema_ptr schema, std::ma
     }
     validate_tags(tags_map);
 
-    std::stringstream serialized_tags;
-    serialized_tags << '{';
-    for (auto& tag_entry : tags_map) {
-        serialized_tags << format("'{}':'{}',", tag_entry.first, tag_entry.second);
-    }
-    std::string serialized_tags_str = serialized_tags.str();
-    if (!tags_map.empty()) {
-        serialized_tags_str[serialized_tags_str.size() - 1] = '}'; // trims the last ',' delimiter
-    } else {
-        serialized_tags_str.push_back('}');
-    }
-
-    sstring req = format("ALTER TABLE \"{}\".\"{}\" WITH {} = {}",
-            schema->ks_name(), schema->cf_name(), tags_extension::NAME, serialized_tags_str);
-    return db::execute_cql(std::move(req)).discard_result();
+    schema_builder builder(schema);
+    builder.set_extensions(schema::extensions_map{{sstring(tags_extension::NAME), ::make_shared<tags_extension>(std::move(tags_map))}});
+    return mm.announce_column_family_update(builder.build(), false, std::vector<view_ptr>(), false);
 }
 
-static future<> add_tags(service::storage_proxy& proxy, schema_ptr schema, rjson::value& request_info) {
+static future<> add_tags(service::migration_manager& mm, service::storage_proxy& proxy, schema_ptr schema, rjson::value& request_info) {
     const rjson::value* tags = rjson::find(request_info, "Tags");
     if (!tags || !tags->IsArray()) {
         return make_exception_future<>(api_error("ValidationException", format("Cannot parse tags")));
@@ -619,7 +607,7 @@ static future<> add_tags(service::storage_proxy& proxy, schema_ptr schema, rjson
     }
 
     std::map<sstring, sstring> tags_map = get_tags_of_table(schema);
-    return update_tags(rjson::copy(*tags), schema, std::move(tags_map), update_tags_action::add_tags);
+    return update_tags(mm, rjson::copy(*tags), schema, std::move(tags_map), update_tags_action::add_tags);
 }
 
 future<executor::request_return_type> executor::tag_resource(client_state& client_state, service_permit permit, rjson::value request) {
@@ -631,7 +619,7 @@ future<executor::request_return_type> executor::tag_resource(client_state& clien
             return api_error("AccessDeniedException", "Incorrect resource identifier");
         }
         schema_ptr schema = get_table_from_arn(_proxy, std::string_view(arn->GetString(), arn->GetStringLength()));
-        add_tags(_proxy, schema, request).get();
+        add_tags(_mm, _proxy, schema, request).get();
         return json_string("");
     });
 }
@@ -652,7 +640,7 @@ future<executor::request_return_type> executor::untag_resource(client_state& cli
         schema_ptr schema = get_table_from_arn(_proxy, std::string_view(arn->GetString(), arn->GetStringLength()));
 
         std::map<sstring, sstring> tags_map = get_tags_of_table(schema);
-        update_tags(*tags, schema, std::move(tags_map), update_tags_action::delete_tags).get();
+        update_tags(_mm, *tags, schema, std::move(tags_map), update_tags_action::delete_tags).get();
         return json_string("");
     });
 }
@@ -869,7 +857,7 @@ future<executor::request_return_type> executor::create_table(client_state& clien
             }).then([this, table_info = std::move(table_info), schema] () mutable {
                 future<> f = make_ready_future<>();
                 if (rjson::find(table_info, "Tags")) {
-                    f = add_tags(_proxy, schema, table_info);
+                    f = add_tags(_mm, _proxy, schema, table_info);
                 }
                 return f.then([table_info = std::move(table_info), schema] () mutable {
                     rjson::value status = rjson::empty_object();
@@ -899,15 +887,24 @@ class attribute_collector {
     void add(bytes&& name, atomic_cell&& cell) {
         collected.emplace(std::move(name), std::move(cell));
     }
+    void add(const bytes& name, atomic_cell&& cell) {
+        collected.emplace(name, std::move(cell));
+    }
 public:
     attribute_collector() : collected(attrs_type()->get_keys_type()->as_less_comparator()) { }
-    void put(bytes&& name, bytes&& val, api::timestamp_type ts) {
-        add(std::move(name), atomic_cell::make_live(*bytes_type, ts, std::move(val), atomic_cell::collection_member::yes));
+    void put(bytes&& name, const bytes& val, api::timestamp_type ts) {
+        add(std::move(name), atomic_cell::make_live(*bytes_type, ts, val, atomic_cell::collection_member::yes));
 
     }
+    void put(const bytes& name, const bytes& val, api::timestamp_type ts) {
+        add(name, atomic_cell::make_live(*bytes_type, ts, val, atomic_cell::collection_member::yes));
+    }
     void del(bytes&& name, api::timestamp_type ts) {
         add(std::move(name), atomic_cell::make_dead(ts, gc_clock::now()));
     }
+    void del(const bytes& name, api::timestamp_type ts) {
+        add(name, atomic_cell::make_dead(ts, gc_clock::now()));
+    }
     collection_mutation_description to_mut() {
         collection_mutation_description ret;
         for (auto&& e : collected) {
@@ -987,7 +984,7 @@ public:
     put_or_delete_item(const rjson::value& item, schema_ptr schema, put_item);
     // put_or_delete_item doesn't keep a reference to schema (so it can be
     // moved between shards for LWT) so it needs to be given again to build():
-    mutation build(schema_ptr schema, api::timestamp_type ts);
+    mutation build(schema_ptr schema, api::timestamp_type ts) const;
     const partition_key& pk() const { return _pk; }
     const clustering_key& ck() const { return _ck; }
 };
@@ -1016,7 +1013,7 @@ put_or_delete_item::put_or_delete_item(const rjson::value& item, schema_ptr sche
     }
 }
 
-mutation put_or_delete_item::build(schema_ptr schema, api::timestamp_type ts) {
+mutation put_or_delete_item::build(schema_ptr schema, api::timestamp_type ts) const {
     mutation m(schema, _pk);
     // If there's no clustering key, a tombstone should be created directly
     // on a partition, not on a clustering row - otherwise it will look like
@@ -1038,7 +1035,7 @@ mutation put_or_delete_item::build(schema_ptr schema, api::timestamp_type ts) {
     for (auto& c : *_cells) {
         const column_definition* cdef = schema->get_column_definition(c.column_name);
         if (!cdef) {
-            attrs_collector.put(std::move(c.column_name), std::move(c.value), ts);
+            attrs_collector.put(c.column_name, c.value, ts);
         } else {
             row.cells().apply(*cdef, atomic_cell::make_live(*cdef->type, ts, std::move(c.value)));
         }
@@ -1338,7 +1335,7 @@ public:
                check_needs_read_before_write(_condition_expression) ||
                _returnvalues == returnvalues::ALL_OLD;
     }
-    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) override {
+    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) const override {
         std::unordered_set<std::string> used_attribute_values;
         std::unordered_set<std::string> used_attribute_names;
         if (!verify_expected(_request, previous_item) ||
@@ -1350,6 +1347,7 @@ public:
             // efficient than throwing an exception.
             return {};
         }
+        _return_attributes = {};
         if (_returnvalues == returnvalues::ALL_OLD && previous_item) {
             // previous_item is supposed to have been created with
             // describe_item(), so has the "Item" attribute:
@@ -1416,7 +1414,7 @@ public:
                 check_needs_read_before_write(_condition_expression) ||
                 _returnvalues == returnvalues::ALL_OLD;
     }
-    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) override {
+    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) const override {
         std::unordered_set<std::string> used_attribute_values;
         std::unordered_set<std::string> used_attribute_names;
         if (!verify_expected(_request, previous_item) ||
@@ -1428,6 +1426,7 @@ public:
             // efficient than throwing an exception.
             return {};
         }
+        _return_attributes = {};
         if (_returnvalues == returnvalues::ALL_OLD && previous_item) {
             rjson::value* item = rjson::find(*previous_item, "Item");
             if (item) {
@@ -1511,7 +1510,7 @@ public:
     virtual ~put_or_delete_item_cas_request() = default;
     virtual std::optional<mutation> apply(query::result& qr, const query::partition_slice& slice, api::timestamp_type ts) override {
         std::optional<mutation> ret;
-        for (put_or_delete_item& mutation_builder : _mutation_builders) {
+        for (const put_or_delete_item& mutation_builder : _mutation_builders) {
             // We assume all these builders have the same partition.
             if (ret) {
                 ret->apply(mutation_builder.build(schema, ts));
@@ -2336,7 +2335,7 @@ public:
 
     update_item_operation(service::storage_proxy& proxy, rjson::value&& request);
     virtual ~update_item_operation() = default;
-    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) override;
+    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) const override;
     bool needs_read_before_write() const;
 };
 
@@ -2400,7 +2399,7 @@ update_item_operation::needs_read_before_write() const {
 }
 
 std::optional<mutation>
-update_item_operation::apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) {
+update_item_operation::apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) const {
     std::unordered_set<std::string> used_attribute_values;
     std::unordered_set<std::string> used_attribute_names;
     if (!verify_expected(_request, previous_item) ||

alternator/rmw_operation.hh

@@ -83,7 +83,11 @@ protected:
     // When _returnvalues != NONE, apply() should store here, in JSON form,
     // the values which are to be returned in the "Attributes" field.
     // The default null JSON means do not return an Attributes field at all.
-    rjson::value _return_attributes;
+    // This field is marked "mutable" so that the const apply() can modify
+    // it (see explanation below), but note that because apply() may be
+    // called more than once, if apply() will sometimes set this field it
+    // must set it (even if just to the default empty value) every time.
+    mutable rjson::value _return_attributes;
 public:
     // The constructor of a rmw_operation subclass should parse the request
     // and try to discover as many input errors as it can before really
@@ -96,7 +100,12 @@ public:
     // conditional expression, apply() should return an empty optional.
     // apply() may throw if it encounters input errors not discovered during
     // the constructor.
-    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) = 0;
+    // apply() may be called more than once in case of contention, so it must
+    // not change the state saved in the object (issue #7218 was caused by
+    // violating this). We mark apply() "const" to let the compiler validate
+    // this for us. The output-only field _return_attributes is marked
+    // "mutable" above so that apply() can still write to it.
+    virtual std::optional<mutation> apply(std::unique_ptr<rjson::value> previous_item, api::timestamp_type ts) const = 0;
     // Convert the above apply() into the signature needed by cas_request:
     virtual std::optional<mutation> apply(query::result& qr, const query::partition_slice& slice, api::timestamp_type ts) override;
     virtual ~rmw_operation() = default;

api/storage_service.cc

@@ -252,8 +252,8 @@ void set_storage_service(http_context& ctx, routes& r) {
                 for (auto cf : column_families) {
                     column_families_vec.push_back(&db.find_column_family(keyspace, cf));
                 }
-                return parallel_for_each(column_families_vec, [&cm] (column_family* cf) {
-                    return cm.perform_cleanup(cf);
+                return parallel_for_each(column_families_vec, [&cm, &db] (column_family* cf) {
+                    return cm.perform_cleanup(db, cf);
                 });
             }).then([]{
                 return make_ready_future<json::json_return_type>(0);

cdc/log.cc

@@ -1135,7 +1135,7 @@ public:
                     if (r.row().deleted_at()) {
                         touched_parts.set<stats::part_type::ROW_DELETE>();
                         cdc_op = operation::row_delete;
-                        if (pirow) {
+                        if (pirow && pikey) {
                             for (const column_definition& column: _schema->regular_columns()) {
                                 assert(pirow->has(column.name_as_text()));
                                 auto& cdef = *_log_schema->get_column_definition(log_data_column_name_bytes(column.name()));

configure.py

@@ -381,6 +381,7 @@ scylla_tests = set([
     'test/boost/view_schema_ckey_test',
     'test/boost/vint_serialization_test',
     'test/boost/virtual_reader_test',
+    'test/boost/stall_free_test',
     'test/manual/ec2_snitch_test',
     'test/manual/gce_snitch_test',
     'test/manual/gossip',
@@ -1265,9 +1266,9 @@ def query_seastar_flags(pc_file, link_static_cxx=False):
     return cflags, libs
 
 for mode in build_modes:
-    seastar_cflags, seastar_libs = query_seastar_flags(pc[mode], link_static_cxx=args.staticcxx)
-    modes[mode]['seastar_cflags'] = seastar_cflags
-    modes[mode]['seastar_libs'] = seastar_libs
+    seastar_pc_cflags, seastar_pc_libs = query_seastar_flags(pc[mode], link_static_cxx=args.staticcxx)
+    modes[mode]['seastar_cflags'] = seastar_pc_cflags
+    modes[mode]['seastar_libs'] = seastar_pc_libs
 
 # We need to use experimental features of the zstd library (to use our own allocators for the (de)compression context),
 # which are available only when the library is linked statically.
@@ -1288,6 +1289,46 @@ def configure_zstd(build_dir, mode):
     os.makedirs(zstd_build_dir, exist_ok=True)
     subprocess.check_call(zstd_cmd, shell=False, cwd=zstd_build_dir)
 
+def configure_abseil(build_dir, mode):
+    abseil_build_dir = os.path.join(build_dir, mode, 'abseil')
+
+    abseil_cflags = seastar_cflags + ' ' + modes[mode]['cxx_ld_flags']
+    cmake_mode = MODE_TO_CMAKE_BUILD_TYPE[mode]
+    abseil_cmake_args = [
+        '-DCMAKE_BUILD_TYPE={}'.format(cmake_mode),
+        '-DCMAKE_INSTALL_PREFIX={}'.format(build_dir + '/inst'), # just to avoid a warning from absl
+        '-DCMAKE_C_COMPILER={}'.format(args.cc),
+        '-DCMAKE_CXX_COMPILER={}'.format(args.cxx),
+        '-DCMAKE_CXX_FLAGS_{}={}'.format(cmake_mode.upper(), abseil_cflags),
+    ]
+
+    abseil_cmd = ['cmake', '-G', 'Ninja', os.path.relpath('abseil', abseil_build_dir)] + abseil_cmake_args
+
+    os.makedirs(abseil_build_dir, exist_ok=True)
+    subprocess.check_call(abseil_cmd, shell=False, cwd=abseil_build_dir)
+
+abseil_libs = ['absl/' + lib for lib in [
+    'container/libabsl_hashtablez_sampler.a',
+    'container/libabsl_raw_hash_set.a',
+    'synchronization/libabsl_synchronization.a',
+    'synchronization/libabsl_graphcycles_internal.a',
+    'debugging/libabsl_stacktrace.a',
+    'debugging/libabsl_symbolize.a',
+    'debugging/libabsl_debugging_internal.a',
+    'debugging/libabsl_demangle_internal.a',
+    'time/libabsl_time.a',
+    'time/libabsl_time_zone.a',
+    'numeric/libabsl_int128.a',
+    'hash/libabsl_city.a',
+    'hash/libabsl_hash.a',
+    'base/libabsl_malloc_internal.a',
+    'base/libabsl_spinlock_wait.a',
+    'base/libabsl_base.a',
+    'base/libabsl_dynamic_annotations.a',
+    'base/libabsl_raw_logging_internal.a',
+    'base/libabsl_exponential_biased.a',
+    'base/libabsl_throw_delegate.a']]
+
 args.user_cflags += " " + pkg_config('jsoncpp', '--cflags')
 args.user_cflags += ' -march=' + args.target
 libs = ' '.join([maybe_static(args.staticyamlcpp, '-lyaml-cpp'), '-latomic', '-llz4', '-lz', '-lsnappy', pkg_config('jsoncpp', '--libs'),
@@ -1316,6 +1357,7 @@ if any(filter(thrift_version.startswith, thrift_boost_versions)):
 for pkg in pkgs:
     args.user_cflags += ' ' + pkg_config(pkg, '--cflags')
     libs += ' ' + pkg_config(pkg, '--libs')
+args.user_cflags += '-I abseil'
 user_cflags = args.user_cflags + ' -fvisibility=hidden'
 user_ldflags = args.user_ldflags + ' -fvisibility=hidden'
 if args.staticcxx:
@@ -1346,6 +1388,9 @@ else:
 for mode in build_modes:
     configure_zstd(outdir, mode)
 
+for mode in build_modes:
+    configure_abseil(outdir, mode)
+
 # configure.py may run automatically from an already-existing build.ninja.
 # If the user interrupts configure.py in the middle, we need build.ninja
 # to remain in a valid state.  So we write our output to a temporary
@@ -1480,6 +1525,8 @@ with open(buildfile_tmp, 'w') as f:
                 objs.extend(['$builddir/' + mode + '/' + artifact for artifact in [
                     'libdeflate/libdeflate.a',
                     'zstd/lib/libzstd.a',
+                ] + [
+                    'abseil/' + x for x in abseil_libs
                 ]])
                 objs.append('$builddir/' + mode + '/gen/utils/gz/crc_combine_table.o')
                 if binary in tests:
@@ -1621,6 +1668,12 @@ with open(buildfile_tmp, 'w') as f:
         f.write('  subdir = build/{mode}/zstd\n'.format(**locals()))
         f.write('  target = libzstd.a\n'.format(**locals()))
 
+        for lib in abseil_libs:
+            f.write('build build/{mode}/abseil/{lib}: ninja\n'.format(**locals()))
+            f.write('  pool = submodule_pool\n')
+            f.write('  subdir = build/{mode}/abseil\n'.format(**locals()))
+            f.write('  target = {lib}\n'.format(**locals()))
+
     mode = 'dev' if 'dev' in modes else modes[0]
     f.write('build checkheaders: phony || {}\n'.format(' '.join(['$builddir/{}/{}.o'.format(mode, hh) for hh in headers])))
 

cql3/functions/aggregate_fcts.cc

@@ -267,10 +267,13 @@ public:
     }
 };
 
-/// The same as `impl_max_function_for' but without knowledge of `Type'.
+/// The same as `impl_max_function_for' but without compile-time dependency on `Type'.
 class impl_max_dynamic_function final : public aggregate_function::aggregate {
+    data_type _io_type;
     opt_bytes _max;
 public:
+    impl_max_dynamic_function(data_type io_type) : _io_type(std::move(io_type)) {}
+
     virtual void reset() override {
         _max = {};
     }
@@ -278,12 +281,11 @@ public:
         return _max.value_or(bytes{});
     }
     virtual void add_input(cql_serialization_format sf, const std::vector<opt_bytes>& values) override {
-        if (!values[0]) {
+        if (values.empty() || !values[0]) {
             return;
         }
-        const auto val = *values[0];
-        if (!_max || *_max < val) {
-            _max = val;
+        if (!_max || _io_type->less(*_max, *values[0])) {
+            _max = values[0];
         }
     }
 };
@@ -298,10 +300,13 @@ public:
 };
 
 class max_dynamic_function final : public native_aggregate_function {
+    data_type _io_type;
 public:
-    max_dynamic_function(data_type io_type) : native_aggregate_function("max", io_type, { io_type }) {}
+    max_dynamic_function(data_type io_type)
+            : native_aggregate_function("max", io_type, { io_type })
+            , _io_type(std::move(io_type)) {}
     virtual std::unique_ptr<aggregate> new_aggregate() override {
-        return std::make_unique<impl_max_dynamic_function>();
+        return std::make_unique<impl_max_dynamic_function>(_io_type);
     }
 };
 
@@ -358,10 +363,13 @@ public:
     }
 };
 
-/// The same as `impl_min_function_for' but without knowledge of `Type'.
+/// The same as `impl_min_function_for' but without compile-time dependency on `Type'.
 class impl_min_dynamic_function final : public aggregate_function::aggregate {
+    data_type _io_type;
     opt_bytes _min;
 public:
+    impl_min_dynamic_function(data_type io_type) : _io_type(std::move(io_type)) {}
+
     virtual void reset() override {
         _min = {};
     }
@@ -369,12 +377,11 @@ public:
         return _min.value_or(bytes{});
     }
     virtual void add_input(cql_serialization_format sf, const std::vector<opt_bytes>& values) override {
-        if (!values[0]) {
+        if (values.empty() || !values[0]) {
             return;
         }
-        const auto val = *values[0];
-        if (!_min || val < *_min) {
-            _min = val;
+        if (!_min || _io_type->less(*values[0], *_min)) {
+            _min = values[0];
         }
     }
 };
@@ -389,10 +396,13 @@ public:
 };
 
 class min_dynamic_function final : public native_aggregate_function {
+    data_type _io_type;
 public:
-    min_dynamic_function(data_type io_type) : native_aggregate_function("min", io_type, { io_type }) {}
+    min_dynamic_function(data_type io_type)
+            : native_aggregate_function("min", io_type, { io_type })
+            , _io_type(std::move(io_type)) {}
     virtual std::unique_ptr<aggregate> new_aggregate() override {
-        return std::make_unique<impl_min_dynamic_function>();
+        return std::make_unique<impl_min_dynamic_function>(_io_type);
     }
 };
 

cql3/functions/castas_fcts.cc

@@ -87,17 +87,14 @@ template<typename ToType>
 std::function<data_value(data_value)> make_castas_fctn_from_decimal_to_float() {
     return [](data_value from) -> data_value {
         auto val_from = value_cast<big_decimal>(from);
-        boost::multiprecision::cpp_int ten(10);
-        boost::multiprecision::cpp_rational r = val_from.unscaled_value();
-        r /= boost::multiprecision::pow(ten, val_from.scale());
-        return static_cast<ToType>(r);
+        return static_cast<ToType>(val_from.as_rational());
     };
 }
 
 static utils::multiprecision_int from_decimal_to_cppint(const data_value& from) {
     const auto& val_from = value_cast<big_decimal>(from);
-    boost::multiprecision::cpp_int ten(10);
-    return boost::multiprecision::cpp_int(val_from.unscaled_value() / boost::multiprecision::pow(ten, val_from.scale()));
+    auto r = val_from.as_rational();
+    return utils::multiprecision_int(numerator(r)/denominator(r));
 }
 
 template<typename ToType>

cql3/lists.cc

@@ -357,7 +357,12 @@ lists::setter_by_uuid::execute(mutation& m, const clustering_key_prefix& prefix,
 
     collection_mutation_description mut;
     mut.cells.reserve(1);
-    mut.cells.emplace_back(to_bytes(*index), params.make_cell(*ltype->value_comparator(), *value, atomic_cell::collection_member::yes));
+
+    if (!value) {
+        mut.cells.emplace_back(to_bytes(*index), params.make_dead_cell());
+    } else {
+        mut.cells.emplace_back(to_bytes(*index), params.make_cell(*ltype->value_comparator(), *value, atomic_cell::collection_member::yes));
+    }
 
     m.set_cell(prefix, column, mut.serialize(*ltype));
 }

cql3/relation.cc

@@ -49,7 +49,7 @@ relation::to_column_definition(const schema& schema, const column_identifier::ra
     auto id = entity.prepare_column_identifier(schema);
     auto def = get_column_definition(schema, *id);
     if (!def || def->is_hidden_from_cql()) {
-        throw exceptions::unrecognized_entity_exception(id, shared_from_this());
+        throw exceptions::unrecognized_entity_exception(*id, to_string());
     }
     return *def;
 }

cql3/restrictions/statement_restrictions.cc

@@ -417,7 +417,7 @@ std::vector<const column_definition*> statement_restrictions::get_column_defs_fo
                     _clustering_columns_restrictions->num_prefix_columns_that_need_not_be_filtered();
             for (auto&& cdef : _clustering_columns_restrictions->get_column_defs()) {
                 ::shared_ptr<single_column_restriction> restr;
-                if (single_pk_restrs) {
+                if (single_ck_restrs) {
                     auto it = single_ck_restrs->restrictions().find(cdef);
                     if (it != single_ck_restrs->restrictions().end()) {
                         restr = dynamic_pointer_cast<single_column_restriction>(it->second);
@@ -624,9 +624,6 @@ bool single_column_restriction::EQ::is_satisfied_by(const schema& schema,
         const row& cells,
         const query_options& options,
         gc_clock::time_point now) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
     auto operand = value(options);
     if (operand) {
         auto cell_value = get_value(schema, key, ckey, cells, now);
@@ -641,9 +638,6 @@ bool single_column_restriction::EQ::is_satisfied_by(const schema& schema,
 }
 
 bool single_column_restriction::EQ::is_satisfied_by(bytes_view data, const query_options& options) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
     auto operand = value(options);
     return operand && _column_def.type->compare(*operand, data) == 0;
 }
@@ -654,9 +648,6 @@ bool single_column_restriction::IN::is_satisfied_by(const schema& schema,
         const row& cells,
         const query_options& options,
         gc_clock::time_point now) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
     auto cell_value = get_value(schema, key, ckey, cells, now);
     if (!cell_value) {
         return false;
@@ -670,9 +661,6 @@ bool single_column_restriction::IN::is_satisfied_by(const schema& schema,
 }
 
 bool single_column_restriction::IN::is_satisfied_by(bytes_view data, const query_options& options) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
     auto operands = values(options);
     return boost::algorithm::any_of(operands, [this, &data] (const bytes_opt& operand) {
         return operand && _column_def.type->compare(*operand, data) == 0;
@@ -697,6 +685,11 @@ static query::range<bytes_view> to_range(const term_slice& slice, const query_op
         extract_bound(statements::bound::END));
 }
 
+static bool contains_without_wraparound(
+        const query::range<bytes_view>& range, bytes_view value, const serialized_tri_compare& cmp) {
+    return !range.is_wrap_around(cmp) && range.contains(value, cmp);
+}
+
 bool single_column_restriction::slice::is_satisfied_by(const schema& schema,
         const partition_key& key,
         const clustering_key_prefix& ckey,
@@ -711,15 +704,14 @@ bool single_column_restriction::slice::is_satisfied_by(const schema& schema,
         return false;
     }
     return cell_value->with_linearized([&] (bytes_view cell_value_bv) {
-        return to_range(_slice, options).contains(cell_value_bv, _column_def.type->as_tri_comparator());
+        return contains_without_wraparound(to_range(_slice, options),
+                cell_value_bv, _column_def.type->as_tri_comparator());
     });
 }
 
 bool single_column_restriction::slice::is_satisfied_by(bytes_view data, const query_options& options) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
-    return to_range(_slice, options).contains(data, _column_def.type->underlying_type()->as_tri_comparator());
+    return contains_without_wraparound(to_range(_slice, options),
+            data, _column_def.type->underlying_type()->as_tri_comparator());
 }
 
 bool single_column_restriction::contains::is_satisfied_by(const schema& schema,
@@ -728,9 +720,6 @@ bool single_column_restriction::contains::is_satisfied_by(const schema& schema,
         const row& cells,
         const query_options& options,
         gc_clock::time_point now) const {
-    if (_column_def.type->is_counter()) {
-        fail(unimplemented::cause::COUNTERS);
-    }
     if (!_column_def.type->is_collection()) {
         return false;
     }

cql3/statements/alter_table_statement.cc

@@ -207,6 +207,9 @@ void alter_table_statement::add_column(const schema& schema, const table& cf, sc
                 "because a collection with the same name and a different type has already been used in the past", column_name));
         }
     }
+    if (type->is_counter() && !schema.is_counter()) {
+        throw exceptions::configuration_exception(format("Cannot add a counter column ({}) in a non counter column family", column_name));
+    }
 
     cfm.with_column(column_name.name(), type, is_static ? column_kind::static_column : column_kind::regular_column);
 
@@ -222,7 +225,7 @@ void alter_table_statement::add_column(const schema& schema, const table& cf, sc
             schema_builder builder(view);
             if (view->view_info()->include_all_columns()) {
                 builder.with_column(column_name.name(), type);
-            } else if (view->view_info()->base_non_pk_columns_in_view_pk().empty()) {
+            } else if (!view->view_info()->has_base_non_pk_columns_in_view_pk()) {
                 db::view::create_virtual_column(builder, column_name.name(), type);
             }
             view_updates.push_back(view_ptr(builder.build()));

cql3/statements/cf_prop_defs.cc

@@ -255,7 +255,9 @@ void cf_prop_defs::apply_to_builder(schema_builder& builder, schema::extensions_
         }
     }
 
-    builder.set_default_time_to_live(gc_clock::duration(get_int(KW_DEFAULT_TIME_TO_LIVE, DEFAULT_DEFAULT_TIME_TO_LIVE)));
+    if (has_property(KW_DEFAULT_TIME_TO_LIVE)) {
+        builder.set_default_time_to_live(gc_clock::duration(get_int(KW_DEFAULT_TIME_TO_LIVE, DEFAULT_DEFAULT_TIME_TO_LIVE)));
+    }
 
     if (has_property(KW_SPECULATIVE_RETRY)) {
         builder.set_speculative_retry(get_string(KW_SPECULATIVE_RETRY, builder.get_speculative_retry().to_sstring()));

cql3/statements/select_statement.cc

@@ -1367,8 +1367,8 @@ select_statement::prepare_restrictions(database& db,
         return ::make_shared<restrictions::statement_restrictions>(db, schema, statement_type::SELECT, std::move(_where_clause), bound_names,
             selection->contains_only_static_columns(), selection->contains_a_collection(), for_view, allow_filtering);
     } catch (const exceptions::unrecognized_entity_exception& e) {
-        if (contains_alias(*e.entity)) {
-            throw exceptions::invalid_request_exception(format("Aliases aren't allowed in the where clause ('{}')", e.relation->to_string()));
+        if (contains_alias(e.entity)) {
+            throw exceptions::invalid_request_exception(format("Aliases aren't allowed in the where clause ('{}')", e.relation_str));
         }
         throw;
     }

database.cc

@@ -1323,7 +1323,7 @@ future<mutation> database::do_apply_counter_update(column_family& cf, const froz
             // counter state for each modified cell...
 
             tracing::trace(trace_state, "Reading counter values from the CF");
-            return counter_write_query(m_schema, cf.as_mutation_source(), m.decorated_key(), slice, trace_state)
+            return counter_write_query(m_schema, cf.as_mutation_source(), m.decorated_key(), slice, trace_state, timeout)
                     .then([this, &cf, &m, m_schema, timeout, trace_state] (auto mopt) {
                 // ...now, that we got existing state of all affected counter
                 // cells we can look for our shard in each of them, increment
@@ -1827,7 +1827,11 @@ future<> database::truncate(const keyspace& ks, column_family& cf, timestamp_fun
                             // TODO: indexes.
                             // Note: since discard_sstables was changed to only count tables owned by this shard,
                             // we can get zero rp back. Changed assert, and ensure we save at least low_mark.
-                            assert(low_mark <= rp || rp == db::replay_position());
+                            // #6995 - the assert below was broken in c2c6c71 and remained so for many years. 
+                            // We nowadays do not flush tables with sstables but autosnapshot=false. This means
+                            // the low_mark assertion does not hold, because we maybe/probably never got around to 
+                            // creating the sstables that would create them.
+                            assert(!should_flush || low_mark <= rp || rp == db::replay_position());
                             rp = std::max(low_mark, rp);
                             return truncate_views(cf, truncated_at, should_flush).then([&cf, truncated_at, rp] {
                                 // save_truncation_record() may actually fail after we cached the truncation time
@@ -2005,9 +2009,10 @@ flat_mutation_reader make_multishard_streaming_reader(distributed<database>& db,
             reader_concurrency_semaphore* semaphore;
         };
         distributed<database>& _db;
+        utils::UUID _table_id;
         std::vector<reader_context> _contexts;
     public:
-        explicit streaming_reader_lifecycle_policy(distributed<database>& db) : _db(db), _contexts(smp::count) {
+        streaming_reader_lifecycle_policy(distributed<database>& db, utils::UUID table_id) : _db(db), _table_id(table_id), _contexts(smp::count) {
         }
         virtual flat_mutation_reader create_reader(
                 schema_ptr schema,
@@ -2036,7 +2041,12 @@ flat_mutation_reader make_multishard_streaming_reader(distributed<database>& db,
             });
         }
         virtual reader_concurrency_semaphore& semaphore() override {
-            return *_contexts[engine().cpu_id()].semaphore;
+            const auto shard = engine().cpu_id();
+            if (!_contexts[shard].semaphore) {
+                auto& cf = _db.local().find_column_family(_table_id);
+                _contexts[shard].semaphore = &cf.streaming_read_concurrency_semaphore();
+            }
+            return *_contexts[shard].semaphore;
         }
     };
     auto ms = mutation_source([&db] (schema_ptr s,
@@ -2047,7 +2057,8 @@ flat_mutation_reader make_multishard_streaming_reader(distributed<database>& db,
             tracing::trace_state_ptr trace_state,
             streamed_mutation::forwarding,
             mutation_reader::forwarding fwd_mr) {
-        return make_multishard_combining_reader(make_shared<streaming_reader_lifecycle_policy>(db), std::move(s), pr, ps, pc,
+        auto table_id = s->id();
+        return make_multishard_combining_reader(make_shared<streaming_reader_lifecycle_policy>(db, table_id), std::move(s), pr, ps, pc,
                 std::move(trace_state), fwd_mr);
     });
     auto&& full_slice = schema->full_slice();

database.hh

@@ -55,6 +55,7 @@
 #include <limits>
 #include <cstddef>
 #include "schema_fwd.hh"
+#include "db/view/view.hh"
 #include "db/schema_features.hh"
 #include "gms/feature.hh"
 #include "timestamp.hh"
@@ -885,7 +886,7 @@ public:
     lw_shared_ptr<const sstable_list> get_sstables_including_compacted_undeleted() const;
     const std::vector<sstables::shared_sstable>& compacted_undeleted_sstables() const;
     std::vector<sstables::shared_sstable> select_sstables(const dht::partition_range& range) const;
-    std::vector<sstables::shared_sstable> candidates_for_compaction() const;
+    std::vector<sstables::shared_sstable> non_staging_sstables() const;
     std::vector<sstables::shared_sstable> sstables_need_rewrite() const;
     size_t sstables_count() const;
     std::vector<uint64_t> sstable_count_per_level() const;
@@ -981,8 +982,9 @@ public:
         return *_config.sstables_manager;
     }
 
+    // Reader's schema must be the same as the base schema of each of the views.
     future<> populate_views(
-            std::vector<view_ptr>,
+            std::vector<db::view::view_and_base>,
             dht::token base_token,
             flat_mutation_reader&&);
 
@@ -998,7 +1000,7 @@ private:
     future<row_locker::lock_holder> do_push_view_replica_updates(const schema_ptr& s, mutation&& m, db::timeout_clock::time_point timeout, mutation_source&& source, const io_priority_class& io_priority) const;
     std::vector<view_ptr> affected_views(const schema_ptr& base, const mutation& update) const;
     future<> generate_and_propagate_view_updates(const schema_ptr& base,
-            std::vector<view_ptr>&& views,
+            std::vector<db::view::view_and_base>&& views,
             mutation&& m,
             flat_mutation_reader_opt existings) const;
 

db/commitlog/commitlog.cc

@@ -520,7 +520,7 @@ public:
             _segment_manager->totals.total_size_on_disk -= size_on_disk();
             _segment_manager->totals.total_size -= (size_on_disk() + _buffer.size_bytes());
             _segment_manager->add_file_to_delete(_file_name, _desc);
-        } else {
+        } else if (_segment_manager->cfg.warn_about_segments_left_on_disk_after_shutdown) {
             clogger.warn("Segment {} is dirty and is left on disk.", *this);
         }
     }
@@ -1293,7 +1293,7 @@ future<db::commitlog::segment_manager::sseg_ptr> db::commitlog::segment_manager:
         auto fut = open_file_dma(filename, flags, opt);
         if (cfg.extensions && !cfg.extensions->commitlog_file_extensions().empty()) {
             for (auto * ext : cfg.extensions->commitlog_file_extensions()) {
-                fut = fut.then([ext, filename, flags](file f) {
+                fut = close_on_failure(std::move(fut), [ext, filename, flags](file f) {
                    return ext->wrap_file(filename, f, flags).then([f](file nf) mutable {
                        return nf ? nf : std::move(f);
                    });

db/commitlog/commitlog.hh

@@ -137,6 +137,7 @@ public:
 
         bool reuse_segments = true;
         bool use_o_dsync = false;
+        bool warn_about_segments_left_on_disk_after_shutdown = true;
 
         const db::extensions * extensions = nullptr;
     };

db/commitlog/commitlog_replayer.cc

@@ -299,7 +299,7 @@ future<> db::commitlog_replayer::impl::process(stats* s, commitlog::buffer_and_r
                 mutation m(cf.schema(), fm.decorated_key(*cf.schema()));
                 converting_mutation_partition_applier v(cm, *cf.schema(), m.partition());
                 fm.partition().accept(cm, v);
-                return do_with(std::move(m), [&db, &cf] (mutation m) {
+                return do_with(std::move(m), [&db, &cf] (const mutation& m) {
                     return db.apply_in_memory(m, cf, db::rp_handle(), db::no_timeout);
                 });
             } else {

db/hints/manager.cc

@@ -224,7 +224,9 @@ future<> manager::end_point_hints_manager::stop(drain should_drain) noexcept {
         with_lock(file_update_mutex(), [this] {
             if (_hints_store_anchor) {
                 hints_store_ptr tmp = std::exchange(_hints_store_anchor, nullptr);
-                return tmp->shutdown().finally([tmp] {});
+                return tmp->shutdown().finally([tmp] {
+                    return tmp->release();
+                }).finally([tmp] {});
             }
             return make_ready_future<>();
         }).handle_exception([&eptr] (auto e) { eptr = std::move(e); }).get();
@@ -326,6 +328,10 @@ future<db::commitlog> manager::end_point_hints_manager::add_store() noexcept {
             // HH doesn't utilize the flow that benefits from reusing segments.
             // Therefore let's simply disable it to avoid any possible confusion.
             cfg.reuse_segments = false;
+            // HH leaves segments on disk after commitlog shutdown, and later reads
+            // them when commitlog is re-created. This is expected to happen regularly
+            // during standard HH workload, so no need to print a warning about it.
+            cfg.warn_about_segments_left_on_disk_after_shutdown = false;
 
             return commitlog::create_commitlog(std::move(cfg)).then([this] (commitlog l) {
                 // add_store() is triggered every time hint files are forcefully flushed to I/O (every hints_flush_period).
@@ -352,7 +358,9 @@ future<> manager::end_point_hints_manager::flush_current_hints() noexcept {
         return futurize_apply([this] {
             return with_lock(file_update_mutex(), [this]() -> future<> {
                 return get_or_load().then([] (hints_store_ptr cptr) {
-                    return cptr->shutdown();
+                    return cptr->shutdown().finally([cptr] {
+                        return cptr->release();
+                    }).finally([cptr] {});
                 }).then([this] {
                     // Un-hold the commitlog object. Since we are under the exclusive _file_update_mutex lock there are no
                     // other hints_store_ptr copies and this would destroy the commitlog shared value.

db/schema_tables.cc

@@ -822,6 +822,14 @@ future<> merge_schema(distributed<service::storage_proxy>& proxy, gms::feature_s
     });
 }
 
+future<> recalculate_schema_version(distributed<service::storage_proxy>& proxy, gms::feature_service& feat) {
+    return merge_lock().then([&proxy, &feat] {
+        return update_schema_version_and_announce(proxy, feat.cluster_schema_features());
+    }).finally([] {
+        return merge_unlock();
+    });
+}
+
 future<> merge_schema(distributed<service::storage_proxy>& proxy, std::vector<mutation> mutations, bool do_flush)
 {
     return merge_lock().then([&proxy, mutations = std::move(mutations), do_flush] () mutable {

db/schema_tables.hh

@@ -170,6 +170,13 @@ future<> merge_schema(distributed<service::storage_proxy>& proxy, gms::feature_s
 
 future<> merge_schema(distributed<service::storage_proxy>& proxy, std::vector<mutation> mutations, bool do_flush);
 
+// Recalculates the local schema version and publishes it in gossip.
+//
+// It is safe to call concurrently with recalculate_schema_version() and merge_schema() in which case it
+// is guaranteed that the schema version we end up with after all calls will reflect the most recent state
+// of feature_service and schema tables.
+future<> recalculate_schema_version(distributed<service::storage_proxy>& proxy, gms::feature_service& feat);
+
 future<std::set<sstring>> merge_keyspaces(distributed<service::storage_proxy>& proxy, schema_result&& before, schema_result&& after);
 
 std::vector<mutation> make_create_keyspace_mutations(lw_shared_ptr<keyspace_metadata> keyspace, api::timestamp_type timestamp, bool with_tables_and_types_and_functions = true);

db/system_keyspace.cc

@@ -203,6 +203,7 @@ schema_ptr batchlog() {
         // operations on resulting CFMetaData:
         //    .compactionStrategyClass(LeveledCompactionStrategy.class);
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        builder.set_wait_for_sync_to_commitlog(true);
        return builder.build(schema_builder::compact_storage::no);
@@ -226,6 +227,7 @@ schema_ptr built_indexes() {
         // comment
         "built column indexes"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::yes);
     }();
@@ -272,6 +274,7 @@ schema_ptr built_indexes() {
         // comment
         "information about the local node"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        builder.remove_column("scylla_cpu_sharding_algorithm");
        builder.remove_column("scylla_nr_shards");
@@ -307,6 +310,7 @@ schema_ptr built_indexes() {
         // comment
         "information about known peers in the cluster"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -331,6 +335,7 @@ schema_ptr built_indexes() {
         // comment
         "events related to peers"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -353,6 +358,7 @@ schema_ptr built_indexes() {
         // comment
         "ranges requested for transfer"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -490,6 +496,7 @@ schema_ptr size_estimates() {
         // comment
         "partitions larger than specified threshold"
         )));
+        builder.set_gc_grace_seconds(0);
         builder.with_version(generate_schema_version(builder.uuid()));
         return builder.build(schema_builder::compact_storage::no);
     }();
@@ -510,6 +517,7 @@ static schema_ptr large_rows() {
                 .with_column("compaction_time", timestamp_type)
                 .set_comment("rows larger than specified threshold")
                 .with_version(generate_schema_version(id))
+                .set_gc_grace_seconds(0)
                 .build();
     }();
     return large_rows;
@@ -530,6 +538,7 @@ static schema_ptr large_cells() {
                 .with_column("compaction_time", timestamp_type)
                 .set_comment("cells larger than specified threshold")
                 .with_version(generate_schema_version(id))
+                .set_gc_grace_seconds(0)
                 .build();
     }();
     return large_cells;
@@ -553,6 +562,7 @@ static schema_ptr large_cells() {
         // comment
         "Scylla specific information about the local node"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -666,6 +676,7 @@ schema_ptr local() {
         // comment
         "information about the local node"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -693,6 +704,7 @@ schema_ptr truncated() {
         // comment
         "information about table truncation"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();
@@ -755,6 +767,7 @@ schema_ptr available_ranges() {
         // comment
         "available keyspace/ranges during bootstrap/replace that are ready to be served"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build();
     }();
@@ -777,6 +790,7 @@ schema_ptr views_builds_in_progress() {
         // comment
         "views builds current progress"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build();
     }();
@@ -799,6 +813,7 @@ schema_ptr built_views() {
         // comment
         "built views"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build();
     }();
@@ -842,6 +857,7 @@ schema_ptr scylla_views_builds_in_progress() {
         // comment
         "CDC-specific information that the local node stores"
        )));
+       builder.set_gc_grace_seconds(0);
        builder.with_version(generate_schema_version(builder.uuid()));
        return builder.build(schema_builder::compact_storage::no);
     }();

db/view/view.cc

@@ -130,17 +130,26 @@ const column_definition* view_info::view_column(const column_definition& base_de
     return _schema.get_column_definition(base_def.name());
 }
 
-const std::vector<column_id>& view_info::base_non_pk_columns_in_view_pk() const {
-    return _base_non_pk_columns_in_view_pk;
+void view_info::set_base_info(db::view::base_info_ptr base_info) {
+    _base_info = std::move(base_info);
 }
 
-void view_info::initialize_base_dependent_fields(const schema& base) {
+db::view::base_info_ptr view_info::make_base_dependent_view_info(const schema& base) const {
+    std::vector<column_id> base_non_pk_columns_in_view_pk;
     for (auto&& view_col : boost::range::join(_schema.partition_key_columns(), _schema.clustering_key_columns())) {
         auto* base_col = base.get_column_definition(view_col.name());
         if (base_col && !base_col->is_primary_key()) {
-            _base_non_pk_columns_in_view_pk.push_back(base_col->id);
+            base_non_pk_columns_in_view_pk.push_back(base_col->id);
         }
     }
+    return make_lw_shared<db::view::base_dependent_view_info>({
+        .base_schema = base.shared_from_this(),
+        .base_non_pk_columns_in_view_pk = std::move(base_non_pk_columns_in_view_pk)
+    });
+}
+
+bool view_info::has_base_non_pk_columns_in_view_pk() const {
+    return !_base_info->base_non_pk_columns_in_view_pk.empty();
 }
 
 namespace db {
@@ -188,11 +197,11 @@ bool may_be_affected_by(const schema& base, const view_info& view, const dht::de
 }
 
 static bool update_requires_read_before_write(const schema& base,
-        const std::vector<view_ptr>& views,
+        const std::vector<view_and_base>& views,
         const dht::decorated_key& key,
         const rows_entry& update) {
     for (auto&& v : views) {
-        view_info& vf = *v->view_info();
+        view_info& vf = *v.view->view_info();
         if (may_be_affected_by(base, vf, key, update)) {
             return true;
         }
@@ -239,12 +248,14 @@ class view_updates final {
     view_ptr _view;
     const view_info& _view_info;
     schema_ptr _base;
+    base_info_ptr _base_info;
     std::unordered_map<partition_key, mutation_partition, partition_key::hashing, partition_key::equality> _updates;
 public:
-    explicit view_updates(view_ptr view, schema_ptr base)
-            : _view(std::move(view))
+    explicit view_updates(view_and_base vab)
+            : _view(std::move(vab.view))
             , _view_info(*_view->view_info())
-            , _base(std::move(base))
+            , _base(vab.base->base_schema)
+            , _base_info(vab.base)
             , _updates(8, partition_key::hashing(*_view), partition_key::equality(*_view)) {
     }
 
@@ -306,7 +317,7 @@ row_marker view_updates::compute_row_marker(const clustering_row& base_row) cons
     // they share liveness information. It's true especially in the only case currently allowed by CQL,
     // which assumes there's up to one non-pk column in the view key. It's also true in alternator,
     // which does not carry TTL information.
-    const auto& col_ids = _view_info.base_non_pk_columns_in_view_pk();
+    const auto& col_ids = _base_info->base_non_pk_columns_in_view_pk;
     if (!col_ids.empty()) {
         auto& def = _base->regular_column_at(col_ids[0]);
         // Note: multi-cell columns can't be part of the primary key.
@@ -537,7 +548,7 @@ void view_updates::delete_old_entry(const partition_key& base_key, const cluster
 
 void view_updates::do_delete_old_entry(const partition_key& base_key, const clustering_row& existing, const clustering_row& update, gc_clock::time_point now) {
     auto& r = get_view_row(base_key, existing);
-    const auto& col_ids = _view_info.base_non_pk_columns_in_view_pk();
+    const auto& col_ids = _base_info->base_non_pk_columns_in_view_pk;
     if (!col_ids.empty()) {
         // We delete the old row using a shadowable row tombstone, making sure that
         // the tombstone deletes everything in the row (or it might still show up).
@@ -678,7 +689,7 @@ void view_updates::generate_update(
         return;
     }
 
-    const auto& col_ids = _view_info.base_non_pk_columns_in_view_pk();
+    const auto& col_ids = _base_info->base_non_pk_columns_in_view_pk;
     if (col_ids.empty()) {
         // The view key is necessarily the same pre and post update.
         if (existing && existing->is_live(*_base)) {
@@ -932,11 +943,16 @@ future<stop_iteration> view_update_builder::on_results() {
 
 future<std::vector<frozen_mutation_and_schema>> generate_view_updates(
         const schema_ptr& base,
-        std::vector<view_ptr>&& views_to_update,
+        std::vector<view_and_base>&& views_to_update,
         flat_mutation_reader&& updates,
         flat_mutation_reader_opt&& existings) {
-    auto vs = boost::copy_range<std::vector<view_updates>>(views_to_update | boost::adaptors::transformed([&] (auto&& v) {
-        return view_updates(std::move(v), base);
+    auto vs = boost::copy_range<std::vector<view_updates>>(views_to_update | boost::adaptors::transformed([&] (view_and_base v) {
+        if (base->version() != v.base->base_schema->version()) {
+            on_internal_error(vlogger, format("Schema version used for view updates ({}) does not match the current"
+                                              " base schema version of the view ({}) for view {}.{} of {}.{}",
+                base->version(), v.base->base_schema->version(), v.view->ks_name(), v.view->cf_name(), base->ks_name(), base->cf_name()));
+        }
+        return view_updates(std::move(v));
     }));
     auto builder = std::make_unique<view_update_builder>(base, std::move(vs), std::move(updates), std::move(existings));
     auto f = builder->build();
@@ -946,18 +962,18 @@ future<std::vector<frozen_mutation_and_schema>> generate_view_updates(
 query::clustering_row_ranges calculate_affected_clustering_ranges(const schema& base,
         const dht::decorated_key& key,
         const mutation_partition& mp,
-        const std::vector<view_ptr>& views) {
+        const std::vector<view_and_base>& views) {
     std::vector<nonwrapping_range<clustering_key_prefix_view>> row_ranges;
     std::vector<nonwrapping_range<clustering_key_prefix_view>> view_row_ranges;
     clustering_key_prefix_view::tri_compare cmp(base);
     if (mp.partition_tombstone() || !mp.row_tombstones().empty()) {
         for (auto&& v : views) {
             // FIXME: #2371
-            if (v->view_info()->select_statement().get_restrictions()->has_unrestricted_clustering_columns()) {
+            if (v.view->view_info()->select_statement().get_restrictions()->has_unrestricted_clustering_columns()) {
                 view_row_ranges.push_back(nonwrapping_range<clustering_key_prefix_view>::make_open_ended_both_sides());
                 break;
             }
-            for (auto&& r : v->view_info()->partition_slice().default_row_ranges()) {
+            for (auto&& r : v.view->view_info()->partition_slice().default_row_ranges()) {
                 view_row_ranges.push_back(r.transform(std::mem_fn(&clustering_key_prefix::view)));
             }
         }
@@ -1717,7 +1733,7 @@ public:
             return stop_iteration::yes;
         }
 
-        _fragments_memory_usage += cr.memory_usage(*_step.base->schema());
+        _fragments_memory_usage += cr.memory_usage(*_step.reader.schema());
         _fragments.push_back(std::move(cr));
         if (_fragments_memory_usage > batch_memory_max) {
             // Although we have not yet completed the batch of base rows that
@@ -1737,10 +1753,14 @@ public:
         _builder._as.check();
         if (!_fragments.empty()) {
             _fragments.push_front(partition_start(_step.current_key, tombstone()));
+            auto base_schema = _step.base->schema();
+            auto views = with_base_info_snapshot(_views_to_build);
+            auto reader = make_flat_mutation_reader_from_fragments(_step.reader.schema(), std::move(_fragments));
+            reader.upgrade_schema(base_schema);
             _step.base->populate_views(
-                    _views_to_build,
+                    std::move(views),
                     _step.current_token(),
-                    make_flat_mutation_reader_from_fragments(_step.base->schema(), std::move(_fragments))).get();
+                    std::move(reader)).get();
             _fragments.clear();
             _fragments_memory_usage = 0;
         }
@@ -1887,5 +1907,11 @@ future<bool> check_needs_view_update_path(db::system_distributed_keyspace& sys_d
     });
 }
 
+std::vector<db::view::view_and_base> with_base_info_snapshot(std::vector<view_ptr> vs) {
+    return boost::copy_range<std::vector<db::view::view_and_base>>(vs | boost::adaptors::transformed([] (const view_ptr& v) {
+        return db::view::view_and_base{v, v->view_info()->base_info()};
+    }));
+}
+
 } // namespace view
 } // namespace db

db/view/view.hh

@@ -43,6 +43,27 @@ namespace db {
 
 namespace view {
 
+// Part of the view description which depends on the base schema version.
+//
+// This structure may change even though the view schema doesn't change, so
+// it needs to live outside view_ptr.
+struct base_dependent_view_info {
+    schema_ptr base_schema;
+
+    // Id of a regular base table column included in the view's PK, if any.
+    // Scylla views only allow one such column, alternator can have up to two.
+    std::vector<column_id> base_non_pk_columns_in_view_pk;
+};
+
+// Immutable snapshot of view's base-schema-dependent part.
+using base_info_ptr = lw_shared_ptr<const base_dependent_view_info>;
+
+// Snapshot of the view schema and its base-schema-dependent part.
+struct view_and_base {
+    view_ptr view;
+    base_info_ptr base;
+};
+
 /**
  * Whether the view filter considers the specified partition key.
  *
@@ -92,7 +113,7 @@ bool clustering_prefix_matches(const schema& base, const partition_key& key, con
 
 future<std::vector<frozen_mutation_and_schema>> generate_view_updates(
         const schema_ptr& base,
-        std::vector<view_ptr>&& views_to_update,
+        std::vector<view_and_base>&& views_to_update,
         flat_mutation_reader&& updates,
         flat_mutation_reader_opt&& existings);
 
@@ -100,7 +121,7 @@ query::clustering_row_ranges calculate_affected_clustering_ranges(
         const schema& base,
         const dht::decorated_key& key,
         const mutation_partition& mp,
-        const std::vector<view_ptr>& views);
+        const std::vector<view_and_base>& views);
 
 struct wait_for_all_updates_tag {};
 using wait_for_all_updates = bool_class<wait_for_all_updates_tag>;
@@ -128,6 +149,13 @@ future<> mutate_MV(
  */
  void create_virtual_column(schema_builder& builder, const bytes& name, const data_type& type);
 
+/**
+ * Converts a collection of view schema snapshots into a collection of
+ * view_and_base objects, which are snapshots of both the view schema
+ * and the base-schema-dependent part of view description.
+ */
+std::vector<view_and_base> with_base_info_snapshot(std::vector<view_ptr>);
+
 }
 
 }

dist/common/scripts/scylla-housekeeping

@@ -61,7 +61,15 @@ def sh_command(*args):
     return out
 
 def get_url(path):
-    return urllib.request.urlopen(path).read().decode('utf-8')
+    # If server returns any error, like 403, or 500 urllib.request throws exception, which is not serializable.
+    # When multiprocessing routines fail to serialize it, it throws ambiguous serialization exception
+    #   from get_json_from_url.
+    # In order to see legit error we catch it from the inside of process, covert to string and
+    #   pass it as part of return value
+    try:
+        return 0, urllib.request.urlopen(path).read().decode('utf-8')
+    except Exception as exc:
+        return 1, str(exc)
 
 def get_json_from_url(path):
     pool = mp.Pool(processes=1)
@@ -71,13 +79,16 @@ def get_json_from_url(path):
     # to enforce a wallclock timeout.
     result = pool.apply_async(get_url, args=(path,))
     try:
-        retval = result.get(timeout=5)
+        status, retval = result.get(timeout=5)
     except mp.TimeoutError as err:
         pool.terminate()
         pool.join()
         raise
+    if status == 1:
+        raise RuntimeError(f'Failed to get "{path}" due to the following error: {retval}')
     return json.loads(retval)
 
+
 def get_api(path):
     return get_json_from_url("http://" + api_address + path)
 

dist/common/scripts/scylla_setup

@@ -371,6 +371,9 @@ if __name__ == '__main__':
                 if not stat.S_ISBLK(os.stat(dsk).st_mode):
                     print('{} is not block device'.format(dsk))
                     continue
+                if dsk in selected:
+                    print(f'{dsk} is already added')
+                    continue
                 selected.append(dsk)
                 devices.remove(dsk)
             disks = ','.join(selected)

dist/common/scripts/scylla_util.py

@@ -182,7 +182,7 @@ class aws_instance:
         instance_size = self.instance_size()
         if instance_class in ['c3', 'c4', 'd2', 'i2', 'r3']:
             return 'ixgbevf'
-        if instance_class in ['a1', 'c5', 'c5d', 'f1', 'g3', 'g4', 'h1', 'i3', 'i3en', 'inf1', 'm5', 'm5a', 'm5ad', 'm5d', 'm5dn', 'm5n', 'm6g', 'p2', 'p3', 'r4', 'r5', 'r5a', 'r5ad', 'r5d', 'r5dn', 'r5n', 't3', 't3a', 'u-6tb1', 'u-9tb1', 'u-12tb1', 'u-18tn1', 'u-24tb1', 'x1', 'x1e', 'z1d']:
+        if instance_class in ['a1', 'c5', 'c5a', 'c5d', 'c5n', 'c6g', 'c6gd', 'f1', 'g3', 'g4', 'h1', 'i3', 'i3en', 'inf1', 'm5', 'm5a', 'm5ad', 'm5d', 'm5dn', 'm5n', 'm6g', 'm6gd', 'p2', 'p3', 'r4', 'r5', 'r5a', 'r5ad', 'r5d', 'r5dn', 'r5n', 't3', 't3a', 'u-6tb1', 'u-9tb1', 'u-12tb1', 'u-18tn1', 'u-24tb1', 'x1', 'x1e', 'z1d']:
             return 'ena'
         if instance_class == 'm4':
             if instance_size == '16xlarge':
@@ -329,7 +329,7 @@ class scylla_cpuinfo:
 
 # When a CLI tool is not installed, use relocatable CLI tool provided by Scylla
 scylla_env = os.environ.copy()
-scylla_env['PATH'] =  '{}:{}'.format(scylla_env['PATH'], scyllabindir())
+scylla_env['PATH'] =  '{}:{}'.format(scyllabindir(), scylla_env['PATH'])
 
 def run(cmd, shell=False, silent=False, exception=True):
     stdout = subprocess.DEVNULL if silent else None
@@ -441,6 +441,19 @@ def dist_ver():
     return platform.dist()[1]
 
 
+SYSTEM_PARTITION_UUIDS = [
+        '21686148-6449-6e6f-744e-656564454649', # BIOS boot partition
+        'c12a7328-f81f-11d2-ba4b-00a0c93ec93b', # EFI system partition
+        '024dee41-33e7-11d3-9d69-0008c781f39f'  # MBR partition scheme
+]
+
+def get_partition_uuid(dev):
+    return out(f'lsblk -n -oPARTTYPE {dev}')
+
+def is_system_partition(dev):
+    uuid = get_partition_uuid(dev)
+    return (uuid in SYSTEM_PARTITION_UUIDS)
+
 def is_unused_disk(dev):
     # dev is not in /sys/class/block/, like /dev/nvme[0-9]+
     if not os.path.isdir('/sys/class/block/{dev}'.format(dev=dev.replace('/dev/', ''))):
@@ -448,7 +461,8 @@ def is_unused_disk(dev):
     try:
         fd = os.open(dev, os.O_EXCL)
         os.close(fd)
-        return True
+        # dev is not reserved for system
+        return not is_system_partition(dev)
     except OSError:
         return False
 

dist/debian/control.mustache

@@ -5,6 +5,7 @@ Section: database
 Priority: optional
 X-Python3-Version: >= 3.4
 Standards-Version: 3.9.5
+Rules-Requires-Root: no
 
 Package: {{product}}-conf
 Architecture: any

dist/debian/python3/control.mustache

@@ -5,6 +5,7 @@ Section: python
 Priority: optional
 X-Python3-Version: >= 3.4
 Standards-Version: 3.9.5
+Rules-Requires-Root: no
 
 Package: {{product}}-python3
 Architecture: amd64

dist/debian/rules.mustache

@@ -37,6 +37,7 @@ override_dh_strip:
 	# The binaries (ethtool...patchelf) don't pass dh_strip after going through patchelf. Since they are
 	# already stripped, nothing is lost if we exclude them, so that's what we do.
 	dh_strip -Xlibprotobuf.so.15 -Xld.so -Xethtool -Xgawk -Xgzip -Xhwloc-calc -Xhwloc-distrib -Xifconfig -Xlscpu -Xnetstat -Xpatchelf --dbg-package={{product}}-server-dbg
+	find $(CURDIR)/debian/{{product}}-server-dbg/usr/lib/debug/.build-id/ -name "*.debug" -exec objcopy --decompress-debug-sections {} \;
 
 override_dh_makeshlibs:
 

exceptions/unrecognized_entity_exception.hh

@@ -56,22 +56,22 @@ public:
     /**
      * The unrecognized entity.
      */
-    ::shared_ptr<cql3::column_identifier> entity;
+    cql3::column_identifier entity;
 
     /**
-     * The entity relation.
+     * The entity relation in a stringified form.
      */
-    cql3::relation_ptr relation;
+    sstring relation_str;
 
     /**
      * Creates a new <code>UnrecognizedEntityException</code>.
      * @param entity the unrecognized entity
-     * @param relation the entity relation
+     * @param relation_str the entity relation string
      */
-    unrecognized_entity_exception(::shared_ptr<cql3::column_identifier> entity, cql3::relation_ptr relation)
-        : invalid_request_exception(format("Undefined name {} in where clause ('{}')", *entity, relation->to_string()))
-        , entity(entity)
-        , relation(relation)
+    unrecognized_entity_exception(cql3::column_identifier entity, sstring relation_str)
+        : invalid_request_exception(format("Undefined name {} in where clause ('{}')", entity, relation_str))
+        , entity(std::move(entity))
+        , relation_str(std::move(relation_str))
     { }
 };
 

flat_mutation_reader.hh

@@ -487,6 +487,9 @@ public:
     size_t buffer_size() const {
         return _impl->buffer_size();
     }
+    const circular_buffer<mutation_fragment>& buffer() const {
+        return _impl->buffer();
+    }
     // Detach the internal buffer of the reader.
     // Roughly equivalent to depleting it by calling pop_mutation_fragment()
     // until is_buffer_empty() returns true.

gms/gossiper.cc

@@ -428,6 +428,7 @@ future<> gossiper::handle_shutdown_msg(inet_address from) {
         return make_ready_future<>();
     }
     return seastar::async([this, from] {
+        auto permit = this->lock_endpoint(from).get0();
         this->mark_as_shutdown(from);
     });
 }
@@ -632,7 +633,7 @@ void gossiper::remove_endpoint(inet_address endpoint) {
     // We can not run on_remove callbacks here becasue on_remove in
     // storage_service might take the gossiper::timer_callback_lock
     (void)seastar::async([this, endpoint] {
-        _subscribers.for_each([endpoint] (auto& subscriber) {
+        _subscribers.for_each([endpoint] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
             subscriber->on_remove(endpoint);
         });
     }).handle_exception([] (auto ep) {
@@ -1464,7 +1465,7 @@ void gossiper::real_mark_alive(inet_address addr, endpoint_state& local_state) {
         logger.info("InetAddress {} is now UP, status = {}", addr, status);
     }
 
-    _subscribers.for_each([addr, local_state] (auto& subscriber) {
+    _subscribers.for_each([addr, local_state] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
         subscriber->on_alive(addr, local_state);
         logger.trace("Notified {}", subscriber.get());
     });
@@ -1478,7 +1479,7 @@ void gossiper::mark_dead(inet_address addr, endpoint_state& local_state) {
     _live_endpoints_just_added.remove(addr);
     _unreachable_endpoints[addr] = now();
     logger.info("InetAddress {} is now DOWN, status = {}", addr, get_gossip_status(local_state));
-    _subscribers.for_each([addr, local_state] (auto& subscriber) {
+    _subscribers.for_each([addr, local_state] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
         subscriber->on_dead(addr, local_state);
         logger.trace("Notified {}", subscriber.get());
     });
@@ -1510,7 +1511,7 @@ void gossiper::handle_major_state_change(inet_address ep, const endpoint_state&
 
     if (eps_old) {
         // the node restarted: it is up to the subscriber to take whatever action is necessary
-        _subscribers.for_each([ep, eps_old] (auto& subscriber) {
+        _subscribers.for_each([ep, eps_old] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
             subscriber->on_restart(ep, *eps_old);
         });
     }
@@ -1525,7 +1526,7 @@ void gossiper::handle_major_state_change(inet_address ep, const endpoint_state&
 
     auto* eps_new = get_endpoint_state_for_endpoint_ptr(ep);
     if (eps_new) {
-        _subscribers.for_each([ep, eps_new] (auto& subscriber) {
+        _subscribers.for_each([ep, eps_new] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
             subscriber->on_join(ep, *eps_new);
         });
     }
@@ -1618,14 +1619,14 @@ void gossiper::apply_new_states(inet_address addr, endpoint_state& local_state,
 
 // Runs inside seastar::async context
 void gossiper::do_before_change_notifications(inet_address addr, const endpoint_state& ep_state, const application_state& ap_state, const versioned_value& new_value) {
-    _subscribers.for_each([addr, ep_state, ap_state, new_value] (auto& subscriber) {
+    _subscribers.for_each([addr, ep_state, ap_state, new_value] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
         subscriber->before_change(addr, ep_state, ap_state, new_value);
     });
 }
 
 // Runs inside seastar::async context
 void gossiper::do_on_change_notifications(inet_address addr, const application_state& state, const versioned_value& value) {
-    _subscribers.for_each([addr, state, value] (auto& subscriber) {
+    _subscribers.for_each([addr, state, value] (shared_ptr<i_endpoint_state_change_subscriber> subscriber) {
         subscriber->on_change(addr, state, value);
     });
 }

install.sh

@@ -126,6 +126,7 @@ relocate_python3() {
     cp "$script" "$relocateddir"
     cat > "$install"<<EOF
 #!/usr/bin/env bash
+export LC_ALL=en_US.UTF-8
 x="\$(readlink -f "\$0")"
 b="\$(basename "\$x")"
 d="\$(dirname "\$x")"

licenses/abseil-license.txt

@@ -0,0 +1,203 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        https://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       https://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+

locator/abstract_replication_strategy.cc

@@ -144,10 +144,33 @@ insert_token_range_to_sorted_container_while_unwrapping(
 
 dht::token_range_vector
 abstract_replication_strategy::get_ranges(inet_address ep) const {
+    return do_get_ranges(ep, _token_metadata, false);
+}
+
+dht::token_range_vector
+abstract_replication_strategy::get_ranges_in_thread(inet_address ep) const {
+    return do_get_ranges(ep, _token_metadata, true);
+}
+
+dht::token_range_vector
+abstract_replication_strategy::get_ranges(inet_address ep, token_metadata& tm) const {
+    return do_get_ranges(ep, tm, false);
+}
+
+dht::token_range_vector
+abstract_replication_strategy::get_ranges_in_thread(inet_address ep, token_metadata& tm) const {
+    return do_get_ranges(ep, tm, true);
+}
+
+dht::token_range_vector
+abstract_replication_strategy::do_get_ranges(inet_address ep, token_metadata& tm, bool can_yield) const {
     dht::token_range_vector ret;
-    auto prev_tok = _token_metadata.sorted_tokens().back();
-    for (auto tok : _token_metadata.sorted_tokens()) {
-        for (inet_address a : calculate_natural_endpoints(tok, _token_metadata)) {
+    auto prev_tok = tm.sorted_tokens().back();
+    for (auto tok : tm.sorted_tokens()) {
+        for (inet_address a : calculate_natural_endpoints(tok, tm)) {
+            if (can_yield) {
+                seastar::thread::maybe_yield();
+            }
             if (a == ep) {
                 insert_token_range_to_sorted_container_while_unwrapping(prev_tok, tok, ret);
                 break;

locator/abstract_replication_strategy.hh

@@ -106,6 +106,15 @@ public:
     // It the analogue of Origin's getAddressRanges().get(endpoint).
     // This function is not efficient, and not meant for the fast path.
     dht::token_range_vector get_ranges(inet_address ep) const;
+    dht::token_range_vector get_ranges_in_thread(inet_address ep) const;
+
+    // Use the token_metadata provided by the caller instead of _token_metadata
+    dht::token_range_vector get_ranges(inet_address ep, token_metadata& tm) const;
+    dht::token_range_vector get_ranges_in_thread(inet_address ep, token_metadata& tm) const;
+private:
+    dht::token_range_vector do_get_ranges(inet_address ep, token_metadata& tm, bool can_yield) const;
+
+public:
     // get_primary_ranges() returns the list of "primary ranges" for the given
     // endpoint. "Primary ranges" are the ranges that the node is responsible
     // for storing replica primarily, which means this is the first node

lua.cc

@@ -264,14 +264,12 @@ static auto visit_lua_raw_value(lua_State* l, int index, Func&& f) {
 
 template <typename Func>
 static auto visit_decimal(const big_decimal &v, Func&& f) {
-    boost::multiprecision::cpp_int ten(10);
-    const auto& dividend = v.unscaled_value();
-    auto divisor = boost::multiprecision::pow(ten, v.scale());
+    boost::multiprecision::cpp_rational r = v.as_rational();
+    const boost::multiprecision::cpp_int& dividend = numerator(r);
+    const boost::multiprecision::cpp_int& divisor = denominator(r);
     if (dividend % divisor == 0) {
-        return f(utils::multiprecision_int(boost::multiprecision::cpp_int(dividend/divisor)));
+        return f(utils::multiprecision_int(dividend/divisor));
     }
-    boost::multiprecision::cpp_rational r = dividend;
-    r /= divisor;
     return f(r.convert_to<double>());
 }
 

main.cc

@@ -546,9 +546,13 @@ int main(int ac, char** av) {
             gms::feature_config fcfg = gms::feature_config_from_db_config(*cfg);
 
             feature_service.start(fcfg).get();
-            auto stop_feature_service = defer_verbose_shutdown("feature service", [&feature_service] {
-                feature_service.stop().get();
-            });
+            // FIXME storage_proxy holds a reference on it and is not yet stopped.
+            // also the proxy leaves range_slice_read_executor-s hanging around
+            // and willing to find out if the cluster_supports_digest_multipartition_reads
+            //
+            //auto stop_feature_service = defer_verbose_shutdown("feature service", [&feature_service] {
+            //    feature_service.stop().get();
+            //});
 
             schema::set_default_partitioner(cfg->partitioner(), cfg->murmur3_partitioner_ignore_msb_bits());
             auto make_sched_group = [&] (sstring name, unsigned shares) {
@@ -943,12 +947,16 @@ int main(int ac, char** av) {
                 mm.init_messaging_service();
             }).get();
             supervisor::notify("initializing storage proxy RPC verbs");
-            proxy.invoke_on_all([] (service::storage_proxy& p) {
-                p.init_messaging_service();
-            }).get();
+            proxy.invoke_on_all(&service::storage_proxy::init_messaging_service).get();
+            auto stop_proxy_handlers = defer_verbose_shutdown("storage proxy RPC verbs", [&proxy] {
+                proxy.invoke_on_all(&service::storage_proxy::uninit_messaging_service).get();
+            });
 
             supervisor::notify("starting streaming service");
             streaming::stream_session::init_streaming_service(db, sys_dist_ks, view_update_generator).get();
+            auto stop_streaming_service = defer_verbose_shutdown("streaming service", [] {
+                streaming::stream_session::uninit_streaming_service().get();
+            });
             api::set_server_stream_manager(ctx).get();
 
             supervisor::notify("starting hinted handoff manager");
@@ -981,6 +989,9 @@ int main(int ac, char** av) {
                 rs.stop().get();
             });
             repair_init_messaging_service_handler(rs, sys_dist_ks, view_update_generator).get();
+            auto stop_repair_messages = defer_verbose_shutdown("repair message handlers", [] {
+                repair_uninit_messaging_service_handler().get();
+            });
             supervisor::notify("starting storage service", true);
             auto& ss = service::get_local_storage_service();
             ss.init_messaging_service_part().get();

message/messaging_service.cc

@@ -718,6 +718,10 @@ void messaging_service::register_stream_mutation_fragments(std::function<future<
     register_handler(this, messaging_verb::STREAM_MUTATION_FRAGMENTS, std::move(func));
 }
 
+future<> messaging_service::unregister_stream_mutation_fragments() {
+    return unregister_handler(messaging_verb::STREAM_MUTATION_FRAGMENTS);
+}
+
 template<class SinkType, class SourceType>
 future<rpc::sink<SinkType>, rpc::source<SourceType>>
 do_make_sink_source(messaging_verb verb, uint32_t repair_meta_id, shared_ptr<messaging_service::rpc_protocol_client_wrapper> rpc_client, std::unique_ptr<messaging_service::rpc_protocol_wrapper>& rpc) {
@@ -749,6 +753,9 @@ rpc::sink<repair_row_on_wire_with_cmd> messaging_service::make_sink_for_repair_g
 void messaging_service::register_repair_get_row_diff_with_rpc_stream(std::function<future<rpc::sink<repair_row_on_wire_with_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_hash_with_cmd> source)>&& func) {
     register_handler(this, messaging_verb::REPAIR_GET_ROW_DIFF_WITH_RPC_STREAM, std::move(func));
 }
+future<> messaging_service::unregister_repair_get_row_diff_with_rpc_stream() {
+    return unregister_handler(messaging_verb::REPAIR_GET_ROW_DIFF_WITH_RPC_STREAM);
+}
 
 // Wrapper for REPAIR_PUT_ROW_DIFF_WITH_RPC_STREAM
 future<rpc::sink<repair_row_on_wire_with_cmd>, rpc::source<repair_stream_cmd>>
@@ -768,6 +775,9 @@ rpc::sink<repair_stream_cmd> messaging_service::make_sink_for_repair_put_row_dif
 void messaging_service::register_repair_put_row_diff_with_rpc_stream(std::function<future<rpc::sink<repair_stream_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_row_on_wire_with_cmd> source)>&& func) {
     register_handler(this, messaging_verb::REPAIR_PUT_ROW_DIFF_WITH_RPC_STREAM, std::move(func));
 }
+future<> messaging_service::unregister_repair_put_row_diff_with_rpc_stream() {
+    return unregister_handler(messaging_verb::REPAIR_PUT_ROW_DIFF_WITH_RPC_STREAM);
+}
 
 // Wrapper for REPAIR_GET_FULL_ROW_HASHES_WITH_RPC_STREAM
 future<rpc::sink<repair_stream_cmd>, rpc::source<repair_hash_with_cmd>>
@@ -787,6 +797,9 @@ rpc::sink<repair_hash_with_cmd> messaging_service::make_sink_for_repair_get_full
 void messaging_service::register_repair_get_full_row_hashes_with_rpc_stream(std::function<future<rpc::sink<repair_hash_with_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_stream_cmd> source)>&& func) {
     register_handler(this, messaging_verb::REPAIR_GET_FULL_ROW_HASHES_WITH_RPC_STREAM, std::move(func));
 }
+future<> messaging_service::unregister_repair_get_full_row_hashes_with_rpc_stream() {
+    return unregister_handler(messaging_verb::REPAIR_GET_FULL_ROW_HASHES_WITH_RPC_STREAM);
+}
 
 // Send a message for verb
 template <typename MsgIn, typename... MsgOut>
@@ -870,6 +883,9 @@ future<streaming::prepare_message> messaging_service::send_prepare_message(msg_a
     return send_message<streaming::prepare_message>(this, messaging_verb::PREPARE_MESSAGE, id,
         std::move(msg), plan_id, std::move(description), reason);
 }
+future<> messaging_service::unregister_prepare_message() {
+    return unregister_handler(messaging_verb::PREPARE_MESSAGE);
+}
 
 // PREPARE_DONE_MESSAGE
 void messaging_service::register_prepare_done_message(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, unsigned dst_cpu_id)>&& func) {
@@ -879,6 +895,9 @@ future<> messaging_service::send_prepare_done_message(msg_addr id, UUID plan_id,
     return send_message<void>(this, messaging_verb::PREPARE_DONE_MESSAGE, id,
         plan_id, dst_cpu_id);
 }
+future<> messaging_service::unregister_prepare_done_message() {
+    return unregister_handler(messaging_verb::PREPARE_DONE_MESSAGE);
+}
 
 // STREAM_MUTATION
 void messaging_service::register_stream_mutation(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, frozen_mutation fm, unsigned dst_cpu_id, rpc::optional<bool> fragmented, rpc::optional<streaming::stream_reason> reason)>&& func) {
@@ -903,6 +922,9 @@ future<> messaging_service::send_stream_mutation_done(msg_addr id, UUID plan_id,
     return send_message<void>(this, messaging_verb::STREAM_MUTATION_DONE, id,
         plan_id, std::move(ranges), cf_id, dst_cpu_id);
 }
+future<> messaging_service::unregister_stream_mutation_done() {
+    return unregister_handler(messaging_verb::STREAM_MUTATION_DONE);
+}
 
 // COMPLETE_MESSAGE
 void messaging_service::register_complete_message(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, unsigned dst_cpu_id, rpc::optional<bool> failed)>&& func) {
@@ -912,6 +934,9 @@ future<> messaging_service::send_complete_message(msg_addr id, UUID plan_id, uns
     return send_message<void>(this, messaging_verb::COMPLETE_MESSAGE, id,
         plan_id, dst_cpu_id, failed);
 }
+future<> messaging_service::unregister_complete_message() {
+    return unregister_handler(messaging_verb::COMPLETE_MESSAGE);
+}
 
 void messaging_service::register_gossip_echo(std::function<future<> ()>&& func) {
     register_handler(this, messaging_verb::GOSSIP_ECHO, std::move(func));
@@ -1126,14 +1151,14 @@ future<partition_checksum> messaging_service::send_repair_checksum_range(
 }
 
 // Wrapper for REPAIR_GET_FULL_ROW_HASHES
-void messaging_service::register_repair_get_full_row_hashes(std::function<future<std::unordered_set<repair_hash>> (const rpc::client_info& cinfo, uint32_t repair_meta_id)>&& func) {
+void messaging_service::register_repair_get_full_row_hashes(std::function<future<repair_hash_set> (const rpc::client_info& cinfo, uint32_t repair_meta_id)>&& func) {
     register_handler(this, messaging_verb::REPAIR_GET_FULL_ROW_HASHES, std::move(func));
 }
 future<> messaging_service::unregister_repair_get_full_row_hashes() {
     return unregister_handler(messaging_verb::REPAIR_GET_FULL_ROW_HASHES);
 }
-future<std::unordered_set<repair_hash>> messaging_service::send_repair_get_full_row_hashes(msg_addr id, uint32_t repair_meta_id) {
-    return send_message<future<std::unordered_set<repair_hash>>>(this, messaging_verb::REPAIR_GET_FULL_ROW_HASHES, std::move(id), repair_meta_id);
+future<repair_hash_set> messaging_service::send_repair_get_full_row_hashes(msg_addr id, uint32_t repair_meta_id) {
+    return send_message<future<repair_hash_set>>(this, messaging_verb::REPAIR_GET_FULL_ROW_HASHES, std::move(id), repair_meta_id);
 }
 
 // Wrapper for REPAIR_GET_COMBINED_ROW_HASH
@@ -1158,13 +1183,13 @@ future<get_sync_boundary_response> messaging_service::send_repair_get_sync_bound
 }
 
 // Wrapper for REPAIR_GET_ROW_DIFF
-void messaging_service::register_repair_get_row_diff(std::function<future<repair_rows_on_wire> (const rpc::client_info& cinfo, uint32_t repair_meta_id, std::unordered_set<repair_hash> set_diff, bool needs_all_rows)>&& func) {
+void messaging_service::register_repair_get_row_diff(std::function<future<repair_rows_on_wire> (const rpc::client_info& cinfo, uint32_t repair_meta_id, repair_hash_set set_diff, bool needs_all_rows)>&& func) {
     register_handler(this, messaging_verb::REPAIR_GET_ROW_DIFF, std::move(func));
 }
 future<> messaging_service::unregister_repair_get_row_diff() {
     return unregister_handler(messaging_verb::REPAIR_GET_ROW_DIFF);
 }
-future<repair_rows_on_wire> messaging_service::send_repair_get_row_diff(msg_addr id, uint32_t repair_meta_id, std::unordered_set<repair_hash> set_diff, bool needs_all_rows) {
+future<repair_rows_on_wire> messaging_service::send_repair_get_row_diff(msg_addr id, uint32_t repair_meta_id, repair_hash_set set_diff, bool needs_all_rows) {
     return send_message<future<repair_rows_on_wire>>(this, messaging_verb::REPAIR_GET_ROW_DIFF, std::move(id), repair_meta_id, std::move(set_diff), needs_all_rows);
 }
 

message/messaging_service.hh

@@ -275,10 +275,12 @@ public:
             streaming::prepare_message msg, UUID plan_id, sstring description, rpc::optional<streaming::stream_reason> reason)>&& func);
     future<streaming::prepare_message> send_prepare_message(msg_addr id, streaming::prepare_message msg, UUID plan_id,
             sstring description, streaming::stream_reason);
+    future<> unregister_prepare_message();
 
     // Wrapper for PREPARE_DONE_MESSAGE verb
     void register_prepare_done_message(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, unsigned dst_cpu_id)>&& func);
     future<> send_prepare_done_message(msg_addr id, UUID plan_id, unsigned dst_cpu_id);
+    future<> unregister_prepare_done_message();
 
     // Wrapper for STREAM_MUTATION verb
     void register_stream_mutation(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, frozen_mutation fm, unsigned dst_cpu_id, rpc::optional<bool>, rpc::optional<streaming::stream_reason>)>&& func);
@@ -287,6 +289,7 @@ public:
     // Wrapper for STREAM_MUTATION_FRAGMENTS
     // The receiver of STREAM_MUTATION_FRAGMENTS sends status code to the sender to notify any error on the receiver side. The status code is of type int32_t. 0 means successful, -1 means error, other status code value are reserved for future use.
     void register_stream_mutation_fragments(std::function<future<rpc::sink<int32_t>> (const rpc::client_info& cinfo, UUID plan_id, UUID schema_id, UUID cf_id, uint64_t estimated_partitions, rpc::optional<streaming::stream_reason> reason_opt, rpc::source<frozen_mutation_fragment, rpc::optional<streaming::stream_mutation_fragments_cmd>> source)>&& func);
+    future<> unregister_stream_mutation_fragments();
     rpc::sink<int32_t> make_sink_for_stream_mutation_fragments(rpc::source<frozen_mutation_fragment, rpc::optional<streaming::stream_mutation_fragments_cmd>>& source);
     future<rpc::sink<frozen_mutation_fragment, streaming::stream_mutation_fragments_cmd>, rpc::source<int32_t>> make_sink_and_source_for_stream_mutation_fragments(utils::UUID schema_id, utils::UUID plan_id, utils::UUID cf_id, uint64_t estimated_partitions, streaming::stream_reason reason, msg_addr id);
 
@@ -294,22 +297,27 @@ public:
     future<rpc::sink<repair_hash_with_cmd>, rpc::source<repair_row_on_wire_with_cmd>> make_sink_and_source_for_repair_get_row_diff_with_rpc_stream(uint32_t repair_meta_id, msg_addr id);
     rpc::sink<repair_row_on_wire_with_cmd> make_sink_for_repair_get_row_diff_with_rpc_stream(rpc::source<repair_hash_with_cmd>& source);
     void register_repair_get_row_diff_with_rpc_stream(std::function<future<rpc::sink<repair_row_on_wire_with_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_hash_with_cmd> source)>&& func);
+    future<> unregister_repair_get_row_diff_with_rpc_stream();
 
     // Wrapper for REPAIR_PUT_ROW_DIFF_WITH_RPC_STREAM
     future<rpc::sink<repair_row_on_wire_with_cmd>, rpc::source<repair_stream_cmd>> make_sink_and_source_for_repair_put_row_diff_with_rpc_stream(uint32_t repair_meta_id, msg_addr id);
     rpc::sink<repair_stream_cmd> make_sink_for_repair_put_row_diff_with_rpc_stream(rpc::source<repair_row_on_wire_with_cmd>& source);
     void register_repair_put_row_diff_with_rpc_stream(std::function<future<rpc::sink<repair_stream_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_row_on_wire_with_cmd> source)>&& func);
+    future<> unregister_repair_put_row_diff_with_rpc_stream();
 
     // Wrapper for REPAIR_GET_FULL_ROW_HASHES_WITH_RPC_STREAM
     future<rpc::sink<repair_stream_cmd>, rpc::source<repair_hash_with_cmd>> make_sink_and_source_for_repair_get_full_row_hashes_with_rpc_stream(uint32_t repair_meta_id, msg_addr id);
     rpc::sink<repair_hash_with_cmd> make_sink_for_repair_get_full_row_hashes_with_rpc_stream(rpc::source<repair_stream_cmd>& source);
     void register_repair_get_full_row_hashes_with_rpc_stream(std::function<future<rpc::sink<repair_hash_with_cmd>> (const rpc::client_info& cinfo, uint32_t repair_meta_id, rpc::source<repair_stream_cmd> source)>&& func);
+    future<> unregister_repair_get_full_row_hashes_with_rpc_stream();
 
     void register_stream_mutation_done(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, dht::token_range_vector ranges, UUID cf_id, unsigned dst_cpu_id)>&& func);
     future<> send_stream_mutation_done(msg_addr id, UUID plan_id, dht::token_range_vector ranges, UUID cf_id, unsigned dst_cpu_id);
+    future<> unregister_stream_mutation_done();
 
     void register_complete_message(std::function<future<> (const rpc::client_info& cinfo, UUID plan_id, unsigned dst_cpu_id, rpc::optional<bool> failed)>&& func);
     future<> send_complete_message(msg_addr id, UUID plan_id, unsigned dst_cpu_id, bool failed = false);
+    future<> unregister_complete_message();
 
     // Wrapper for REPAIR_CHECKSUM_RANGE verb
     void register_repair_checksum_range(std::function<future<partition_checksum> (sstring keyspace, sstring cf, dht::token_range range, rpc::optional<repair_checksum> hash_version)>&& func);
@@ -317,9 +325,9 @@ public:
     future<partition_checksum> send_repair_checksum_range(msg_addr id, sstring keyspace, sstring cf, dht::token_range range, repair_checksum hash_version);
 
     // Wrapper for REPAIR_GET_FULL_ROW_HASHES
-    void register_repair_get_full_row_hashes(std::function<future<std::unordered_set<repair_hash>> (const rpc::client_info& cinfo, uint32_t repair_meta_id)>&& func);
+    void register_repair_get_full_row_hashes(std::function<future<repair_hash_set> (const rpc::client_info& cinfo, uint32_t repair_meta_id)>&& func);
     future<> unregister_repair_get_full_row_hashes();
-    future<std::unordered_set<repair_hash>> send_repair_get_full_row_hashes(msg_addr id, uint32_t repair_meta_id);
+    future<repair_hash_set> send_repair_get_full_row_hashes(msg_addr id, uint32_t repair_meta_id);
 
     // Wrapper for REPAIR_GET_COMBINED_ROW_HASH
     void register_repair_get_combined_row_hash(std::function<future<get_combined_row_hash_response> (const rpc::client_info& cinfo, uint32_t repair_meta_id, std::optional<repair_sync_boundary> common_sync_boundary)>&& func);
@@ -332,9 +340,9 @@ public:
     future<get_sync_boundary_response> send_repair_get_sync_boundary(msg_addr id, uint32_t repair_meta_id, std::optional<repair_sync_boundary> skipped_sync_boundary);
 
     // Wrapper for REPAIR_GET_ROW_DIFF
-    void register_repair_get_row_diff(std::function<future<repair_rows_on_wire> (const rpc::client_info& cinfo, uint32_t repair_meta_id, std::unordered_set<repair_hash> set_diff, bool needs_all_rows)>&& func);
+    void register_repair_get_row_diff(std::function<future<repair_rows_on_wire> (const rpc::client_info& cinfo, uint32_t repair_meta_id, repair_hash_set set_diff, bool needs_all_rows)>&& func);
     future<> unregister_repair_get_row_diff();
-    future<repair_rows_on_wire> send_repair_get_row_diff(msg_addr id, uint32_t repair_meta_id, std::unordered_set<repair_hash> set_diff, bool needs_all_rows);
+    future<repair_rows_on_wire> send_repair_get_row_diff(msg_addr id, uint32_t repair_meta_id, repair_hash_set set_diff, bool needs_all_rows);
 
     // Wrapper for REPAIR_PUT_ROW_DIFF
     void register_repair_put_row_diff(std::function<future<> (const rpc::client_info& cinfo, uint32_t repair_meta_id, repair_rows_on_wire row_diff)>&& func);

multishard_mutation_query.cc

@@ -195,6 +195,7 @@ class read_context : public reader_lifecycle_policy {
 
     // One for each shard. Index is shard id.
     std::vector<reader_meta> _readers;
+    std::vector<reader_concurrency_semaphore*> _semaphores;
 
     gate _dismantling_gate;
 
@@ -211,7 +212,8 @@ public:
             , _schema(std::move(s))
             , _cmd(cmd)
             , _ranges(ranges)
-            , _trace_state(std::move(trace_state)) {
+            , _trace_state(std::move(trace_state))
+            , _semaphores(smp::count, nullptr) {
         _readers.resize(smp::count);
     }
 
@@ -236,7 +238,12 @@ public:
     virtual void destroy_reader(shard_id shard, future<stopped_reader> reader_fut) noexcept override;
 
     virtual reader_concurrency_semaphore& semaphore() override {
-        return _readers[engine().cpu_id()].rparts->semaphore;
+        const auto shard = engine().cpu_id();
+        if (!_semaphores[shard]) {
+            auto& table = _db.local().find_column_family(_schema);
+            _semaphores[shard] = &table.read_concurrency_semaphore();
+        }
+        return *_semaphores[shard];
     }
 
     future<> lookup_readers();

mutation_partition.cc

@@ -1721,7 +1721,7 @@ void row::apply_monotonically(const schema& s, column_kind kind, row&& other) {
 // we erase the live cells according to the shadowable_tombstone rules.
 static bool dead_marker_shadows_row(const schema& s, column_kind kind, const row_marker& marker) {
     return s.is_view()
-            && !s.view_info()->base_non_pk_columns_in_view_pk().empty()
+            && s.view_info()->has_base_non_pk_columns_in_view_pk()
             && !marker.is_live()
             && kind == column_kind::regular_column; // not applicable to static rows
 }
@@ -2505,7 +2505,8 @@ mutation_partition::fully_discontinuous(const schema& s, const position_range& r
 future<mutation_opt> counter_write_query(schema_ptr s, const mutation_source& source,
                                          const dht::decorated_key& dk,
                                          const query::partition_slice& slice,
-                                         tracing::trace_state_ptr trace_ptr)
+                                         tracing::trace_state_ptr trace_ptr,
+                                         db::timeout_clock::time_point timeout)
 {
     struct range_and_reader {
         dht::partition_range range;
@@ -2530,7 +2531,7 @@ future<mutation_opt> counter_write_query(schema_ptr s, const mutation_source& so
     auto cwqrb = counter_write_query_result_builder(*s);
     auto cfq = make_stable_flattened_mutations_consumer<compact_for_query<emit_only_live_rows::yes, counter_write_query_result_builder>>(
             *s, gc_clock::now(), slice, query::max_rows, query::max_rows, std::move(cwqrb));
-    auto f = r_a_r->reader.consume(std::move(cfq), db::no_timeout);
+    auto f = r_a_r->reader.consume(std::move(cfq), timeout);
     return f.finally([r_a_r = std::move(r_a_r)] { });
 }
 

mutation_query.hh

@@ -113,9 +113,6 @@ class reconcilable_result_builder {
     const schema& _schema;
     const query::partition_slice& _slice;
 
-    utils::chunked_vector<partition> _result;
-    uint32_t _live_rows{};
-
     bool _return_static_content_on_partition_with_no_rows{};
     bool _static_row_is_alive{};
     uint32_t _total_live_rows = 0;
@@ -123,6 +120,10 @@ class reconcilable_result_builder {
     stop_iteration _stop;
     bool _short_read_allowed;
     std::optional<streamed_mutation_freezer> _mutation_consumer;
+
+    uint32_t _live_rows{};
+    // make this the last member so it is destroyed first. #7240
+    utils::chunked_vector<partition> _result;
 public:
     reconcilable_result_builder(const schema& s, const query::partition_slice& slice,
                                 query::result_memory_accounter&& accounter)
@@ -206,5 +207,6 @@ public:
 future<mutation_opt> counter_write_query(schema_ptr, const mutation_source&,
                                          const dht::decorated_key& dk,
                                          const query::partition_slice& slice,
-                                         tracing::trace_state_ptr trace_ptr);
+                                         tracing::trace_state_ptr trace_ptr,
+                                         db::timeout_clock::time_point timeout);
 

mutation_reader.hh

@@ -372,6 +372,64 @@ flat_mutation_reader make_foreign_reader(schema_ptr schema,
         foreign_ptr<std::unique_ptr<flat_mutation_reader>> reader,
         streamed_mutation::forwarding fwd_sm = streamed_mutation::forwarding::no);
 
+/// Make an auto-paused evictable reader.
+///
+/// The reader is paused after each use, that is after each call to any of its
+/// members that cause actual reading to be done (`fill_buffer()` and
+/// `fast_forward_to()`). When paused, the reader is made evictable, that it is
+/// it is registered with reader concurrency semaphore as an inactive read.
+/// The reader is resumed automatically on the next use. If it was evicted, it
+/// will be recreated at the position it left off reading. This is all
+/// transparent to its user.
+/// Parameters passed by reference have to be kept alive while the reader is
+/// alive.
+flat_mutation_reader make_auto_paused_evictable_reader(
+        mutation_source ms,
+        schema_ptr schema,
+        reader_concurrency_semaphore& semaphore,
+        const dht::partition_range& pr,
+        const query::partition_slice& ps,
+        const io_priority_class& pc,
+        tracing::trace_state_ptr trace_state,
+        mutation_reader::forwarding fwd_mr);
+
+class evictable_reader;
+
+class evictable_reader_handle {
+    friend std::pair<flat_mutation_reader, evictable_reader_handle> make_manually_paused_evictable_reader(mutation_source, schema_ptr, reader_concurrency_semaphore&,
+            const dht::partition_range&, const query::partition_slice&, const io_priority_class&, tracing::trace_state_ptr, mutation_reader::forwarding);
+
+private:
+    evictable_reader* _r;
+
+private:
+    explicit evictable_reader_handle(evictable_reader& r);
+
+public:
+    void pause();
+};
+
+/// Make a manually-paused evictable reader.
+///
+/// The reader can be paused via the evictable reader handle when desired. The
+/// intended usage is subsequent reads done in bursts, after which the reader is
+/// not used for some time. When paused, the reader is made evictable, that is,
+/// it is registered with reader concurrency semaphore as an inactive read.
+/// The reader is resumed automatically on the next use. If it was evicted, it
+/// will be recreated at the position it left off reading. This is all
+/// transparent to its user.
+/// Parameters passed by reference have to be kept alive while the reader is
+/// alive.
+std::pair<flat_mutation_reader, evictable_reader_handle> make_manually_paused_evictable_reader(
+        mutation_source ms,
+        schema_ptr schema,
+        reader_concurrency_semaphore& semaphore,
+        const dht::partition_range& pr,
+        const query::partition_slice& ps,
+        const io_priority_class& pc,
+        tracing::trace_state_ptr trace_state,
+        mutation_reader::forwarding fwd_mr);
+
 /// Reader lifecycle policy for the mulitshard combining reader.
 ///
 /// This policy is expected to make sure any additional resource the readers

position_in_partition.hh

@@ -163,6 +163,11 @@ public:
         return {partition_region::clustered, bound_weight::before_all_prefixed, &ck};
     }
 
+    // Returns a view to before_key(pos._ck) if pos.is_clustering_row() else returns pos as-is.
+    static position_in_partition_view before_key(position_in_partition_view pos) {
+        return {partition_region::clustered, pos._bound_weight == bound_weight::equal ? bound_weight::before_all_prefixed : pos._bound_weight, pos._ck};
+    }
+
     partition_region region() const { return _type; }
     bound_weight get_bound_weight() const { return _bound_weight; }
     bool is_partition_start() const { return _type == partition_region::partition_start; }

reader_concurrency_semaphore.cc

@@ -27,6 +27,7 @@
 
 
 reader_permit::impl::impl(reader_concurrency_semaphore& semaphore, reader_resources base_cost) : semaphore(semaphore), base_cost(base_cost) {
+    semaphore.consume(base_cost);
 }
 
 reader_permit::impl::~impl() {
@@ -88,7 +89,6 @@ void reader_concurrency_semaphore::signal(const resources& r) noexcept {
     _resources += r;
     while (!_wait_list.empty() && has_available_units(_wait_list.front().res)) {
         auto& x = _wait_list.front();
-        _resources -= x.res;
         try {
             x.pr.set_value(reader_permit(*this, x.res));
         } catch (...) {
@@ -160,7 +160,6 @@ future<reader_permit> reader_concurrency_semaphore::wait_admission(size_t memory
         --_inactive_read_stats.population;
     }
     if (may_proceed(r)) {
-        _resources -= r;
         return make_ready_future<reader_permit>(reader_permit(*this, r));
     }
     promise<reader_permit> pr;
@@ -170,7 +169,6 @@ future<reader_permit> reader_concurrency_semaphore::wait_admission(size_t memory
 }
 
 reader_permit reader_concurrency_semaphore::consume_resources(resources r) {
-    _resources -= r;
     return reader_permit(*this, r);
 }
 

reader_concurrency_semaphore.hh

@@ -128,6 +128,10 @@ private:
         return has_available_units(r) && _wait_list.empty();
     }
 
+    void consume(resources r) {
+        _resources -= r;
+    }
+
     void consume_memory(size_t memory) {
         _resources.memory -= memory;
     }

repair/repair.hh

@@ -23,6 +23,7 @@
 
 #include <unordered_map>
 #include <exception>
+#include <absl/container/btree_set.h>
 
 #include <seastar/core/sstring.hh>
 #include <seastar/core/sharded.hh>
@@ -334,6 +335,8 @@ public:
     }
 };
 
+using repair_hash_set = absl::btree_set<repair_hash>;
+
 // Return value of the REPAIR_GET_SYNC_BOUNDARY RPC verb
 struct get_sync_boundary_response {
     std::optional<repair_sync_boundary> boundary;

repair/row_level.cc

@@ -47,6 +47,7 @@
 #include "gms/gossiper.hh"
 #include "repair/row_level.hh"
 #include "mutation_source_metadata.hh"
+#include "utils/stall_free.hh"
 
 extern logging::logger rlogger;
 
@@ -373,6 +374,7 @@ private:
     std::optional<utils::phased_barrier::operation> _local_read_op;
     // Local reader or multishard reader to read the range
     flat_mutation_reader _reader;
+    std::optional<evictable_reader_handle> _reader_handle;
     // Current partition read from disk
     lw_shared_ptr<const decorated_key_with_hash> _current_dk;
 
@@ -392,32 +394,49 @@ public:
             , _sharder(remote_partitioner, range, remote_shard)
             , _seed(seed)
             , _local_read_op(local_reader ? std::optional(cf.read_in_progress()) : std::nullopt)
-            , _reader(make_reader(db, cf, local_reader)) {
-    }
-
-private:
-    flat_mutation_reader
-    make_reader(seastar::sharded<database>& db,
-            column_family& cf,
-            is_local_reader local_reader) {
+            , _reader(nullptr) {
         if (local_reader) {
-            return cf.make_streaming_reader(_schema, _range);
+            auto ms = mutation_source([&cf] (
+                        schema_ptr s,
+                        reader_permit,
+                        const dht::partition_range& pr,
+                        const query::partition_slice& ps,
+                        const io_priority_class& pc,
+                        tracing::trace_state_ptr,
+                        streamed_mutation::forwarding,
+                        mutation_reader::forwarding fwd_mr) {
+                return cf.make_streaming_reader(std::move(s), pr, ps, fwd_mr);
+            });
+            std::tie(_reader, _reader_handle) = make_manually_paused_evictable_reader(
+                    std::move(ms),
+                    _schema,
+                    cf.streaming_read_concurrency_semaphore(),
+                    _range,
+                    _schema->full_slice(),
+                    service::get_local_streaming_read_priority(),
+                    {},
+                    mutation_reader::forwarding::no);
+        } else {
+            _reader = make_multishard_streaming_reader(db, _schema, [this] {
+                auto shard_range = _sharder.next();
+                if (shard_range) {
+                    return std::optional<dht::partition_range>(dht::to_partition_range(*shard_range));
+                }
+                return std::optional<dht::partition_range>();
+            });
         }
-        return make_multishard_streaming_reader(db, _schema, [this] {
-            auto shard_range = _sharder.next();
-            if (shard_range) {
-                return std::optional<dht::partition_range>(dht::to_partition_range(*shard_range));
-            }
-            return std::optional<dht::partition_range>();
-        });
     }
 
-public:
     future<mutation_fragment_opt>
     read_mutation_fragment() {
         return _reader(db::no_timeout);
     }
 
+    void on_end_of_stream() {
+        _reader = make_empty_flat_reader(_schema);
+        _reader_handle.reset();
+    }
+
     lw_shared_ptr<const decorated_key_with_hash>& get_current_dk() {
         return _current_dk;
     }
@@ -436,6 +455,11 @@ public:
         }
     }
 
+    void pause() {
+        if (_reader_handle) {
+            _reader_handle->pause();
+        }
+    }
 };
 
 class repair_writer {
@@ -644,7 +668,7 @@ private:
     // Tracks current sync boundary
     std::optional<repair_sync_boundary> _current_sync_boundary;
     // Contains the hashes of rows in the _working_row_buffor for all peer nodes
-    std::vector<std::unordered_set<repair_hash>> _peer_row_hash_sets;
+    std::vector<repair_hash_set> _peer_row_hash_sets;
     // Gate used to make sure pending operation of meta data is done
     seastar::gate _gate;
     sink_source_for_get_full_row_hashes _sink_source_for_get_full_row_hashes;
@@ -733,11 +757,12 @@ public:
 public:
     future<> stop() {
         auto gate_future = _gate.close();
-        auto writer_future = _repair_writer.wait_for_writer_done();
         auto f1 = _sink_source_for_get_full_row_hashes.close();
         auto f2 = _sink_source_for_get_row_diff.close();
         auto f3 = _sink_source_for_put_row_diff.close();
-        return when_all_succeed(std::move(gate_future), std::move(writer_future), std::move(f1), std::move(f2), std::move(f3));
+        return when_all_succeed(std::move(gate_future), std::move(f1), std::move(f2), std::move(f3)).finally([this] {
+            return _repair_writer.wait_for_writer_done();
+        });
     }
 
     static std::unordered_map<node_repair_meta_id, lw_shared_ptr<repair_meta>>& repair_meta_map() {
@@ -865,9 +890,9 @@ public:
     }
 
     // Must run inside a seastar thread
-    static std::unordered_set<repair_hash>
-    get_set_diff(const std::unordered_set<repair_hash>& x, const std::unordered_set<repair_hash>& y) {
-        std::unordered_set<repair_hash> set_diff;
+    static repair_hash_set
+    get_set_diff(const repair_hash_set& x, const repair_hash_set& y) {
+        repair_hash_set set_diff;
         // Note std::set_difference needs x and y are sorted.
         std::copy_if(x.begin(), x.end(), std::inserter(set_diff, set_diff.end()),
                 [&y] (auto& item) { thread::maybe_yield(); return y.find(item) == y.end(); });
@@ -885,14 +910,14 @@ public:
 
     }
 
-    std::unordered_set<repair_hash>& peer_row_hash_sets(unsigned node_idx) {
+    repair_hash_set& peer_row_hash_sets(unsigned node_idx) {
         return _peer_row_hash_sets[node_idx];
     }
 
     // Get a list of row hashes in _working_row_buf
-    future<std::unordered_set<repair_hash>>
+    future<repair_hash_set>
     working_row_hashes() {
-        return do_with(std::unordered_set<repair_hash>(), [this] (std::unordered_set<repair_hash>& hashes) {
+        return do_with(repair_hash_set(), [this] (repair_hash_set& hashes) {
             return do_for_each(_working_row_buf, [&hashes] (repair_row& r) {
                 hashes.emplace(r.hash());
             }).then([&hashes] {
@@ -1018,11 +1043,7 @@ private:
         return repair_hash(h.finalize_uint64());
     }
 
-    stop_iteration handle_mutation_fragment(mutation_fragment_opt mfopt, size_t& cur_size, size_t& new_rows_size, std::list<repair_row>& cur_rows) {
-        if (!mfopt) {
-            return stop_iteration::yes;
-        }
-        mutation_fragment& mf = *mfopt;
+    stop_iteration handle_mutation_fragment(mutation_fragment& mf, size_t& cur_size, size_t& new_rows_size, std::list<repair_row>& cur_rows) {
         if (mf.is_partition_start()) {
             auto& start = mf.as_partition_start();
             _repair_reader.set_current_dk(start.key());
@@ -1057,32 +1078,49 @@ private:
                 }
                 _gate.check();
                 return _repair_reader.read_mutation_fragment().then([this, &cur_size, &new_rows_size, &cur_rows] (mutation_fragment_opt mfopt) mutable {
-                    return handle_mutation_fragment(std::move(mfopt), cur_size, new_rows_size, cur_rows);
+                    if (!mfopt) {
+                        _repair_reader.on_end_of_stream();
+                        return stop_iteration::yes;
+                    }
+                    return handle_mutation_fragment(*mfopt, cur_size, new_rows_size, cur_rows);
                 });
-            }).then([&cur_rows, &new_rows_size] () mutable {
+            }).then_wrapped([this, &cur_rows, &new_rows_size] (future<> fut) mutable {
+                if (fut.failed()) {
+                    _repair_reader.on_end_of_stream();
+                    return make_exception_future<std::list<repair_row>, size_t>(fut.get_exception());
+                }
+                _repair_reader.pause();
                 return make_ready_future<std::list<repair_row>, size_t>(std::move(cur_rows), new_rows_size);
             });
         });
     }
 
+    future<> clear_row_buf() {
+        return utils::clear_gently(_row_buf);
+    }
+
+    future<> clear_working_row_buf() {
+        return utils::clear_gently(_working_row_buf).then([this] {
+            _working_row_buf_combined_hash.clear();
+        });
+    }
+
     // Read rows from disk until _max_row_buf_size of rows are filled into _row_buf.
     // Calculate the combined checksum of the rows
     // Calculate the total size of the rows in _row_buf
     future<get_sync_boundary_response>
     get_sync_boundary(std::optional<repair_sync_boundary> skipped_sync_boundary) {
+        auto f = make_ready_future<>();
         if (skipped_sync_boundary) {
             _current_sync_boundary = skipped_sync_boundary;
-            _row_buf.clear();
-            _working_row_buf.clear();
-            _working_row_buf_combined_hash.clear();
-        } else {
-            _working_row_buf.clear();
-            _working_row_buf_combined_hash.clear();
+            f = clear_row_buf();
         }
         // Here is the place we update _last_sync_boundary
         rlogger.trace("SET _last_sync_boundary from {} to {}", _last_sync_boundary, _current_sync_boundary);
         _last_sync_boundary = _current_sync_boundary;
-        return row_buf_size().then([this, sb = std::move(skipped_sync_boundary)] (size_t cur_size) {
+      return f.then([this, sb = std::move(skipped_sync_boundary)] () mutable {
+       return clear_working_row_buf().then([this, sb = sb] () mutable {
+        return row_buf_size().then([this, sb = std::move(sb)] (size_t cur_size) {
             return read_rows_from_disk(cur_size).then([this, sb = std::move(sb)] (std::list<repair_row> new_rows, size_t new_rows_size) mutable {
                 size_t new_rows_nr = new_rows.size();
                 _row_buf.splice(_row_buf.end(), new_rows);
@@ -1099,6 +1137,8 @@ private:
                 });
             });
         });
+       });
+      });
     }
 
     future<> move_row_buf_to_working_row_buf() {
@@ -1174,9 +1214,9 @@ private:
     }
 
     future<std::list<repair_row>>
-    copy_rows_from_working_row_buf_within_set_diff(std::unordered_set<repair_hash> set_diff) {
+    copy_rows_from_working_row_buf_within_set_diff(repair_hash_set set_diff) {
         return do_with(std::list<repair_row>(), std::move(set_diff),
-                [this] (std::list<repair_row>& rows, std::unordered_set<repair_hash>& set_diff) {
+                [this] (std::list<repair_row>& rows, repair_hash_set& set_diff) {
             return do_for_each(_working_row_buf, [this, &set_diff, &rows] (const repair_row& r) {
                 if (set_diff.count(r.hash()) > 0) {
                     rows.push_back(r);
@@ -1191,7 +1231,7 @@ private:
     // Give a set of row hashes, return the corresponding rows
     // If needs_all_rows is set, return all the rows in _working_row_buf, ignore the set_diff
     future<std::list<repair_row>>
-    get_row_diff(std::unordered_set<repair_hash> set_diff, needs_all_rows_t needs_all_rows = needs_all_rows_t::no) {
+    get_row_diff(repair_hash_set set_diff, needs_all_rows_t needs_all_rows = needs_all_rows_t::no) {
         if (needs_all_rows) {
             if (!_repair_master || _nr_peer_nodes == 1) {
                 return make_ready_future<std::list<repair_row>>(std::move(_working_row_buf));
@@ -1202,19 +1242,28 @@ private:
         }
     }
 
-    future<> do_apply_rows(std::list<repair_row>& row_diff, unsigned node_idx, update_working_row_buf update_buf) {
-        return with_semaphore(_repair_writer.sem(), 1, [this, node_idx, update_buf, &row_diff] {
-            _repair_writer.create_writer(_db, node_idx);
-            return do_for_each(row_diff, [this, node_idx, update_buf] (repair_row& r) {
-                if (update_buf) {
-                    _working_row_buf_combined_hash.add(r.hash());
-                }
-                // The repair_row here is supposed to have
-                // mutation_fragment attached because we have stored it in
-                // to_repair_rows_list above where the repair_row is created.
-                mutation_fragment mf = std::move(r.get_mutation_fragment());
-                auto dk_with_hash = r.get_dk_with_hash();
-                return _repair_writer.do_write(node_idx, std::move(dk_with_hash), std::move(mf));
+    future<> do_apply_rows(std::list<repair_row>&& row_diff, unsigned node_idx, update_working_row_buf update_buf) {
+        return do_with(std::move(row_diff), [this, node_idx, update_buf] (std::list<repair_row>& row_diff) {
+            return with_semaphore(_repair_writer.sem(), 1, [this, node_idx, update_buf, &row_diff] {
+                _repair_writer.create_writer(_db, node_idx);
+                return repeat([this, node_idx, update_buf, &row_diff] () mutable {
+                    if (row_diff.empty()) {
+                        return make_ready_future<stop_iteration>(stop_iteration::yes);
+                    }
+                    repair_row& r = row_diff.front();
+                    if (update_buf) {
+                        _working_row_buf_combined_hash.add(r.hash());
+                    }
+                    // The repair_row here is supposed to have
+                    // mutation_fragment attached because we have stored it in
+                    // to_repair_rows_list above where the repair_row is created.
+                    mutation_fragment mf = std::move(r.get_mutation_fragment());
+                    auto dk_with_hash = r.get_dk_with_hash();
+                    return _repair_writer.do_write(node_idx, std::move(dk_with_hash), std::move(mf)).then([&row_diff] {
+                        row_diff.pop_front();
+                        return make_ready_future<stop_iteration>(stop_iteration::no);
+                    });
+                });
             });
         });
     }
@@ -1232,19 +1281,17 @@ private:
         stats().rx_row_nr += row_diff.size();
         stats().rx_row_nr_peer[from] += row_diff.size();
         if (update_buf) {
-            std::list<repair_row> tmp;
-            tmp.swap(_working_row_buf);
             // Both row_diff and _working_row_buf and are ordered, merging
             // two sored list to make sure the combination of row_diff
             // and _working_row_buf are ordered.
-            std::merge(tmp.begin(), tmp.end(), row_diff.begin(), row_diff.end(), std::back_inserter(_working_row_buf),
-                [this] (const repair_row& x, const repair_row& y) { thread::maybe_yield(); return _cmp(x.boundary(), y.boundary()) < 0; });
+            utils::merge_to_gently(_working_row_buf, row_diff,
+                 [this] (const repair_row& x, const repair_row& y) { return _cmp(x.boundary(), y.boundary()) < 0; });
         }
         if (update_hash_set) {
-            _peer_row_hash_sets[node_idx] = boost::copy_range<std::unordered_set<repair_hash>>(row_diff |
+            _peer_row_hash_sets[node_idx] = boost::copy_range<repair_hash_set>(row_diff |
                     boost::adaptors::transformed([] (repair_row& r) { thread::maybe_yield(); return r.hash(); }));
         }
-        do_apply_rows(row_diff, node_idx, update_buf).get();
+        do_apply_rows(std::move(row_diff), node_idx, update_buf).get();
     }
 
     future<>
@@ -1252,11 +1299,9 @@ private:
         if (rows.empty()) {
             return make_ready_future<>();
         }
-        return to_repair_rows_list(rows).then([this] (std::list<repair_row> row_diff) {
-            return do_with(std::move(row_diff), [this] (std::list<repair_row>& row_diff) {
-                unsigned node_idx = 0;
-                return do_apply_rows(row_diff, node_idx, update_working_row_buf::no);
-            });
+        return to_repair_rows_list(std::move(rows)).then([this] (std::list<repair_row> row_diff) {
+            unsigned node_idx = 0;
+            return do_apply_rows(std::move(row_diff), node_idx, update_working_row_buf::no);
         });
     }
 
@@ -1335,13 +1380,13 @@ private:
 public:
     // RPC API
     // Return the hashes of the rows in _working_row_buf
-    future<std::unordered_set<repair_hash>>
+    future<repair_hash_set>
     get_full_row_hashes(gms::inet_address remote_node) {
         if (remote_node == _myip) {
             return get_full_row_hashes_handler();
         }
         return netw::get_local_messaging_service().send_repair_get_full_row_hashes(msg_addr(remote_node),
-                _repair_meta_id).then([this, remote_node] (std::unordered_set<repair_hash> hashes) {
+                _repair_meta_id).then([this, remote_node] (repair_hash_set hashes) {
             rlogger.debug("Got full hashes from peer={}, nr_hashes={}", remote_node, hashes.size());
             _metrics.rx_hashes_nr += hashes.size();
             stats().rx_hashes_nr += hashes.size();
@@ -1352,7 +1397,7 @@ public:
 
 private:
     future<> get_full_row_hashes_source_op(
-            lw_shared_ptr<std::unordered_set<repair_hash>> current_hashes,
+            lw_shared_ptr<repair_hash_set> current_hashes,
             gms::inet_address remote_node,
             unsigned node_idx,
             rpc::source<repair_hash_with_cmd>& source) {
@@ -1390,12 +1435,12 @@ private:
     }
 
 public:
-    future<std::unordered_set<repair_hash>>
+    future<repair_hash_set>
     get_full_row_hashes_with_rpc_stream(gms::inet_address remote_node, unsigned node_idx) {
         if (remote_node == _myip) {
             return get_full_row_hashes_handler();
         }
-        auto current_hashes = make_lw_shared<std::unordered_set<repair_hash>>();
+        auto current_hashes = make_lw_shared<repair_hash_set>();
         return _sink_source_for_get_full_row_hashes.get_sink_source(remote_node, node_idx).then(
                 [this, current_hashes, remote_node, node_idx]
                 (rpc::sink<repair_stream_cmd>& sink, rpc::source<repair_hash_with_cmd>& source) mutable {
@@ -1410,7 +1455,7 @@ public:
     }
 
     // RPC handler
-    future<std::unordered_set<repair_hash>>
+    future<repair_hash_set>
     get_full_row_hashes_handler() {
         return with_gate(_gate, [this] {
             return working_row_hashes();
@@ -1543,7 +1588,7 @@ public:
     // RPC API
     // Return rows in the _working_row_buf with hash within the given sef_diff
     // Must run inside a seastar thread
-    void get_row_diff(std::unordered_set<repair_hash> set_diff, needs_all_rows_t needs_all_rows, gms::inet_address remote_node, unsigned node_idx) {
+    void get_row_diff(repair_hash_set set_diff, needs_all_rows_t needs_all_rows, gms::inet_address remote_node, unsigned node_idx) {
         if (needs_all_rows || !set_diff.empty()) {
             if (remote_node == _myip) {
                 return;
@@ -1612,11 +1657,11 @@ private:
     }
 
     future<> get_row_diff_sink_op(
-            std::unordered_set<repair_hash> set_diff,
+            repair_hash_set set_diff,
             needs_all_rows_t needs_all_rows,
             rpc::sink<repair_hash_with_cmd>& sink,
             gms::inet_address remote_node) {
-        return do_with(std::move(set_diff), [needs_all_rows, remote_node, &sink] (std::unordered_set<repair_hash>& set_diff) mutable {
+        return do_with(std::move(set_diff), [needs_all_rows, remote_node, &sink] (repair_hash_set& set_diff) mutable {
             if (inject_rpc_stream_error) {
                 return make_exception_future<>(std::runtime_error("get_row_diff: Inject sender error in sink loop"));
             }
@@ -1643,7 +1688,7 @@ private:
 public:
     // Must run inside a seastar thread
     void get_row_diff_with_rpc_stream(
-            std::unordered_set<repair_hash> set_diff,
+            repair_hash_set set_diff,
             needs_all_rows_t needs_all_rows,
             update_peer_row_hash_sets update_hash_set,
             gms::inet_address remote_node,
@@ -1669,7 +1714,7 @@ public:
     }
 
     // RPC handler
-    future<repair_rows_on_wire> get_row_diff_handler(std::unordered_set<repair_hash> set_diff, needs_all_rows_t needs_all_rows) {
+    future<repair_rows_on_wire> get_row_diff_handler(repair_hash_set set_diff, needs_all_rows_t needs_all_rows) {
         return with_gate(_gate, [this, set_diff = std::move(set_diff), needs_all_rows] () mutable {
             return get_row_diff(std::move(set_diff), needs_all_rows).then([this] (std::list<repair_row> row_diff) {
                 return to_repair_rows_on_wire(std::move(row_diff));
@@ -1679,15 +1724,16 @@ public:
 
     // RPC API
     // Send rows in the _working_row_buf with hash within the given sef_diff
-    future<> put_row_diff(std::unordered_set<repair_hash> set_diff, needs_all_rows_t needs_all_rows, gms::inet_address remote_node) {
+    future<> put_row_diff(repair_hash_set set_diff, needs_all_rows_t needs_all_rows, gms::inet_address remote_node) {
         if (!set_diff.empty()) {
             if (remote_node == _myip) {
                 return make_ready_future<>();
             }
-            auto sz = set_diff.size();
+            size_t sz = set_diff.size();
             return get_row_diff(std::move(set_diff), needs_all_rows).then([this, remote_node, sz] (std::list<repair_row> row_diff) {
                 if (row_diff.size() != sz) {
-                    throw std::runtime_error("row_diff.size() != set_diff.size()");
+                    rlogger.warn("Hash conflict detected, keyspace={}, table={}, range={}, row_diff.size={}, set_diff.size={}. It is recommended to compact the table and rerun repair for the range.",
+                            _schema->ks_name(), _schema->cf_name(), _range, row_diff.size(), sz);
                 }
                 return do_with(std::move(row_diff), [this, remote_node] (std::list<repair_row>& row_diff) {
                     return get_repair_rows_size(row_diff).then([this, remote_node, &row_diff] (size_t row_bytes) mutable {
@@ -1754,17 +1800,18 @@ private:
 
 public:
     future<> put_row_diff_with_rpc_stream(
-            std::unordered_set<repair_hash> set_diff,
+            repair_hash_set set_diff,
             needs_all_rows_t needs_all_rows,
             gms::inet_address remote_node, unsigned node_idx) {
         if (!set_diff.empty()) {
             if (remote_node == _myip) {
                 return make_ready_future<>();
             }
-            auto sz = set_diff.size();
+            size_t sz = set_diff.size();
             return get_row_diff(std::move(set_diff), needs_all_rows).then([this, remote_node, node_idx, sz] (std::list<repair_row> row_diff) {
                 if (row_diff.size() != sz) {
-                    throw std::runtime_error("row_diff.size() != set_diff.size()");
+                    rlogger.warn("Hash conflict detected, keyspace={}, table={}, range={}, row_diff.size={}, set_diff.size={}. It is recommended to compact the table and rerun repair for the range.",
+                            _schema->ks_name(), _schema->cf_name(), _range, row_diff.size(), sz);
                 }
                 return do_with(std::move(row_diff), [this, remote_node, node_idx] (std::list<repair_row>& row_diff) {
                     return get_repair_rows_size(row_diff).then([this, remote_node, node_idx, &row_diff] (size_t row_bytes) mutable {
@@ -1803,7 +1850,7 @@ static future<stop_iteration> repair_get_row_diff_with_rpc_stream_process_op(
         rpc::sink<repair_row_on_wire_with_cmd> sink,
         rpc::source<repair_hash_with_cmd> source,
         bool &error,
-        std::unordered_set<repair_hash>& current_set_diff,
+        repair_hash_set& current_set_diff,
         std::optional<std::tuple<repair_hash_with_cmd>> hash_cmd_opt) {
     repair_hash_with_cmd hash_cmd = std::get<0>(hash_cmd_opt.value());
     rlogger.trace("Got repair_hash_with_cmd from peer={}, hash={}, cmd={}", from, hash_cmd.hash, int(hash_cmd.cmd));
@@ -1816,7 +1863,7 @@ static future<stop_iteration> repair_get_row_diff_with_rpc_stream_process_op(
         }
         bool needs_all_rows = hash_cmd.cmd == repair_stream_cmd::needs_all_rows;
         _metrics.rx_hashes_nr += current_set_diff.size();
-        auto fp = make_foreign(std::make_unique<std::unordered_set<repair_hash>>(std::move(current_set_diff)));
+        auto fp = make_foreign(std::make_unique<repair_hash_set>(std::move(current_set_diff)));
         return smp::submit_to(src_cpu_id % smp::count, [from, repair_meta_id, needs_all_rows, fp = std::move(fp)] {
             auto rm = repair_meta::get_repair_meta(from, repair_meta_id);
             if (fp.get_owner_shard() == engine().cpu_id()) {
@@ -1894,12 +1941,12 @@ static future<stop_iteration> repair_get_full_row_hashes_with_rpc_stream_process
     if (status == repair_stream_cmd::get_full_row_hashes) {
         return smp::submit_to(src_cpu_id % smp::count, [from, repair_meta_id] {
             auto rm = repair_meta::get_repair_meta(from, repair_meta_id);
-            return rm->get_full_row_hashes_handler().then([] (std::unordered_set<repair_hash> hashes) {
+            return rm->get_full_row_hashes_handler().then([] (repair_hash_set hashes) {
                 _metrics.tx_hashes_nr += hashes.size();
                 return hashes;
             });
-        }).then([sink] (std::unordered_set<repair_hash> hashes) mutable {
-            return do_with(std::move(hashes), [sink] (std::unordered_set<repair_hash>& hashes) mutable {
+        }).then([sink] (repair_hash_set hashes) mutable {
+            return do_with(std::move(hashes), [sink] (repair_hash_set& hashes) mutable {
                 return do_for_each(hashes, [sink] (const repair_hash& hash) mutable {
                     return sink(repair_hash_with_cmd{repair_stream_cmd::hash_data, hash});
                 }).then([sink] () mutable {
@@ -1922,7 +1969,7 @@ static future<> repair_get_row_diff_with_rpc_stream_handler(
         uint32_t repair_meta_id,
         rpc::sink<repair_row_on_wire_with_cmd> sink,
         rpc::source<repair_hash_with_cmd> source) {
-    return do_with(false, std::unordered_set<repair_hash>(), [from, src_cpu_id, repair_meta_id, sink, source] (bool& error, std::unordered_set<repair_hash>& current_set_diff) mutable {
+    return do_with(false, repair_hash_set(), [from, src_cpu_id, repair_meta_id, sink, source] (bool& error, repair_hash_set& current_set_diff) mutable {
         return repeat([from, src_cpu_id, repair_meta_id, sink, source, &error, &current_set_diff] () mutable {
             return source().then([from, src_cpu_id, repair_meta_id, sink, source, &error, &current_set_diff] (std::optional<std::tuple<repair_hash_with_cmd>> hash_cmd_opt) mutable {
                 if (hash_cmd_opt) {
@@ -2065,7 +2112,7 @@ future<> repair_init_messaging_service_handler(repair_service& rs, distributed<d
             auto from = cinfo.retrieve_auxiliary<gms::inet_address>("baddr");
             return smp::submit_to(src_cpu_id % smp::count, [from, repair_meta_id] {
                 auto rm = repair_meta::get_repair_meta(from, repair_meta_id);
-                return rm->get_full_row_hashes_handler().then([] (std::unordered_set<repair_hash> hashes) {
+                return rm->get_full_row_hashes_handler().then([] (repair_hash_set hashes) {
                     _metrics.tx_hashes_nr += hashes.size();
                     return hashes;
                 });
@@ -2093,11 +2140,11 @@ future<> repair_init_messaging_service_handler(repair_service& rs, distributed<d
             });
         });
         ms.register_repair_get_row_diff([] (const rpc::client_info& cinfo, uint32_t repair_meta_id,
-                std::unordered_set<repair_hash> set_diff, bool needs_all_rows) {
+                repair_hash_set set_diff, bool needs_all_rows) {
             auto src_cpu_id = cinfo.retrieve_auxiliary<uint32_t>("src_cpu_id");
             auto from = cinfo.retrieve_auxiliary<gms::inet_address>("baddr");
             _metrics.rx_hashes_nr += set_diff.size();
-            auto fp = make_foreign(std::make_unique<std::unordered_set<repair_hash>>(std::move(set_diff)));
+            auto fp = make_foreign(std::make_unique<repair_hash_set>(std::move(set_diff)));
             return smp::submit_to(src_cpu_id % smp::count, [from, repair_meta_id, fp = std::move(fp), needs_all_rows] () mutable {
                 auto rm = repair_meta::get_repair_meta(from, repair_meta_id);
                 if (fp.get_owner_shard() == engine().cpu_id()) {
@@ -2165,6 +2212,25 @@ future<> repair_init_messaging_service_handler(repair_service& rs, distributed<d
     });
 }
 
+future<> repair_uninit_messaging_service_handler() {
+    return netw::get_messaging_service().invoke_on_all([] (auto& ms) {
+        return when_all_succeed(
+            ms.unregister_repair_get_row_diff_with_rpc_stream(),
+            ms.unregister_repair_put_row_diff_with_rpc_stream(),
+            ms.unregister_repair_get_full_row_hashes_with_rpc_stream(),
+            ms.unregister_repair_get_full_row_hashes(),
+            ms.unregister_repair_get_combined_row_hash(),
+            ms.unregister_repair_get_sync_boundary(),
+            ms.unregister_repair_get_row_diff(),
+            ms.unregister_repair_put_row_diff(),
+            ms.unregister_repair_row_level_start(),
+            ms.unregister_repair_row_level_stop(),
+            ms.unregister_repair_get_estimated_partitions(),
+            ms.unregister_repair_set_estimated_partitions(),
+            ms.unregister_repair_get_diff_algorithms()).discard_result();
+    });
+}
+
 class row_level_repair {
     repair_info& _ri;
     sstring _cf_name;
@@ -2394,7 +2460,7 @@ private:
             // sequentially because the rows from repair follower 1 to
             // repair master might reduce the amount of missing data
             // between repair master and repair follower 2.
-            std::unordered_set<repair_hash> set_diff = repair_meta::get_set_diff(master.peer_row_hash_sets(node_idx), master.working_row_hashes().get0());
+            repair_hash_set set_diff = repair_meta::get_set_diff(master.peer_row_hash_sets(node_idx), master.working_row_hashes().get0());
             // Request missing sets from peer node
             rlogger.debug("Before get_row_diff to node {}, local={}, peer={}, set_diff={}",
                     node, master.working_row_hashes().get0().size(), master.peer_row_hash_sets(node_idx).size(), set_diff.size());
@@ -2417,9 +2483,9 @@ private:
         // So we can figure out which rows peer node are missing and send the missing rows to them
         check_in_shutdown();
         _ri.check_in_abort();
-        std::unordered_set<repair_hash> local_row_hash_sets = master.working_row_hashes().get0();
+        repair_hash_set local_row_hash_sets = master.working_row_hashes().get0();
         auto sz = _all_live_peer_nodes.size();
-        std::vector<std::unordered_set<repair_hash>> set_diffs(sz);
+        std::vector<repair_hash_set> set_diffs(sz);
         for (size_t idx : boost::irange(size_t(0), sz)) {
             set_diffs[idx] = repair_meta::get_set_diff(local_row_hash_sets, master.peer_row_hash_sets(idx));
         }

repair/row_level.hh

@@ -45,6 +45,7 @@ private:
 };
 
 future<> repair_init_messaging_service_handler(repair_service& rs, distributed<db::system_distributed_keyspace>& sys_dist_ks, distributed<db::view::view_update_generator>& view_update_generator);
+future<> repair_uninit_messaging_service_handler();
 
 class repair_info;
 

schema.cc

@@ -42,6 +42,8 @@
 
 constexpr int32_t schema::NAME_LENGTH;
 
+extern logging::logger dblog;
+
 sstring to_sstring(column_kind k) {
     switch (k) {
     case column_kind::partition_key:  return "PARTITION_KEY";
@@ -575,11 +577,15 @@ schema::get_column_definition(const bytes& name) const {
 
 const column_definition&
 schema::column_at(column_kind kind, column_id id) const {
-    return _raw._columns.at(column_offset(kind) + id);
+    return column_at(static_cast<ordinal_column_id>(column_offset(kind) + id));
 }
 
 const column_definition&
 schema::column_at(ordinal_column_id ordinal_id) const {
+    if (size_t(ordinal_id) >= _raw._columns.size()) {
+        on_internal_error(dblog, format("{}.{}@{}: column id {:d} >= {:d}",
+            ks_name(), cf_name(), version(), size_t(ordinal_id), _raw._columns.size()));
+    }
     return _raw._columns.at(static_cast<column_count_type>(ordinal_id));
 }
 

scripts/create-relocatable-package.py

@@ -79,7 +79,8 @@ executables = ['build/{}/scylla'.format(args.mode),
                '/usr/sbin/ethtool',
                '/usr/bin/netstat',
                '/usr/bin/hwloc-distrib',
-               '/usr/bin/hwloc-calc']
+               '/usr/bin/hwloc-calc',
+               '/usr/bin/lsblk']
 
 output = args.dest
 

scylla-gdb.py

@@ -596,7 +596,7 @@ def current_shard():
 
 
 def find_db(shard=None):
-    if not shard:
+    if shard is None:
         shard = current_shard()
     return gdb.parse_and_eval('::debug::db')['_instances']['_M_impl']['_M_start'][shard]['service']['_p']
 

scylla_post_install.sh

@@ -63,6 +63,17 @@ MemoryHigh=1200M
 MemoryMax=1400M
 MemoryLimit=1400M
 EOS
+
+# On CentOS7, systemd does not support percentage-based parameter.
+# To apply memory parameter on CentOS7, we need to override the parameter
+# in bytes, instead of percentage.
+elif [ "$RHEL" -a "$VERSION_ID" = "7" ]; then
+    MEMORY_LIMIT=$((MEMTOTAL_BYTES / 100 * 5))
+    mkdir -p /etc/systemd/system/scylla-helper.slice.d/
+    cat << EOS > /etc/systemd/system/scylla-helper.slice.d/memory.conf
+[Slice]
+MemoryLimit=$MEMORY_LIMIT
+EOS
 fi
 
 systemctl --system daemon-reload >/dev/null || true

serializer_impl.hh

@@ -25,6 +25,7 @@
 #include <seastar/util/bool_class.hh>
 #include <boost/range/algorithm/for_each.hpp>
 #include "utils/small_vector.hh"
+#include <absl/container/btree_set.h>
 
 namespace ser {
 
@@ -81,6 +82,17 @@ static inline void serialize_array(Output& out, const Container& v) {
 template<typename Container>
 struct container_traits;
 
+template<typename T>
+struct container_traits<absl::btree_set<T>> {
+    struct back_emplacer {
+        absl::btree_set<T>& c;
+        back_emplacer(absl::btree_set<T>& c_) : c(c_) {}
+        void operator()(T&& v) {
+            c.emplace(std::move(v));
+        }
+    };
+};
+
 template<typename T>
 struct container_traits<std::unordered_set<T>> {
     struct back_emplacer {
@@ -253,6 +265,27 @@ struct serializer<std::list<T>> {
     }
 };
 
+template<typename T>
+struct serializer<absl::btree_set<T>> {
+    template<typename Input>
+    static absl::btree_set<T> read(Input& in) {
+        auto sz = deserialize(in, boost::type<uint32_t>());
+        absl::btree_set<T> v;
+        deserialize_array_helper<false, T>::doit(in, v, sz);
+        return v;
+    }
+    template<typename Output>
+    static void write(Output& out, const absl::btree_set<T>& v) {
+        safe_serialize_as_uint32(out, v.size());
+        serialize_array_helper<false, T>::doit(out, v);
+    }
+    template<typename Input>
+    static void skip(Input& in) {
+        auto sz = deserialize(in, boost::type<uint32_t>());
+        skip_array<T>(in, sz);
+    }
+};
+
 template<typename T>
 struct serializer<std::unordered_set<T>> {
     template<typename Input>

service/migration_manager.cc

@@ -92,7 +92,7 @@ void migration_manager::init_messaging_service()
         //FIXME: future discarded.
         (void)with_gate(_background_tasks, [this] {
             mlogger.debug("features changed, recalculating schema version");
-            return update_schema_version_and_announce(get_storage_proxy(), _feat.cluster_schema_features());
+            return db::schema_tables::recalculate_schema_version(get_storage_proxy(), _feat);
         });
     };
 
@@ -277,9 +277,9 @@ future<> migration_manager::maybe_schedule_schema_pull(const utils::UUID& their_
     }).finally([me = shared_from_this()] {});
 }
 
-future<> migration_manager::submit_migration_task(const gms::inet_address& endpoint)
+future<> migration_manager::submit_migration_task(const gms::inet_address& endpoint, bool can_ignore_down_node)
 {
-    return service::migration_task::run_may_throw(endpoint);
+    return service::migration_task::run_may_throw(endpoint, can_ignore_down_node);
 }
 
 future<> migration_manager::do_merge_schema_from(netw::messaging_service::msg_addr id)
@@ -1132,7 +1132,8 @@ future<> migration_manager::sync_schema(const database& db, const std::vector<gm
         }).then([this, &schema_map] {
             return parallel_for_each(schema_map, [this] (auto& x) {
                 mlogger.debug("Pulling schema {} from {}", x.first, x.second.front());
-                return submit_migration_task(x.second.front());
+                bool can_ignore_down_node = false;
+                return submit_migration_task(x.second.front(), can_ignore_down_node);
             });
         });
     });

service/migration_manager.hh

@@ -82,7 +82,7 @@ public:
 
     future<> maybe_schedule_schema_pull(const utils::UUID& their_version, const gms::inet_address& endpoint);
 
-    future<> submit_migration_task(const gms::inet_address& endpoint);
+    future<> submit_migration_task(const gms::inet_address& endpoint, bool can_ignore_down_node = true);
 
     // Makes sure that this node knows about all schema changes known by "nodes" that were made prior to this call.
     future<> sync_schema(const database& db, const std::vector<gms::inet_address>& nodes);

service/migration_task.cc

@@ -51,11 +51,12 @@ namespace service {
 
 static logging::logger mlogger("migration_task");
 
-future<> migration_task::run_may_throw(const gms::inet_address& endpoint)
+future<> migration_task::run_may_throw(const gms::inet_address& endpoint, bool can_ignore_down_node)
 {
     if (!gms::get_local_gossiper().is_alive(endpoint)) {
-        mlogger.warn("Can't send migration request: node {} is down.", endpoint);
-        return make_ready_future<>();
+        auto msg = format("Can't send migration request: node {} is down.", endpoint);
+        mlogger.warn("{}", msg);
+        return can_ignore_down_node ? make_ready_future<>() : make_exception_future<>(std::runtime_error(msg));
     }
     netw::messaging_service::msg_addr id{endpoint, 0};
     return service::get_local_migration_manager().merge_schema_from(id).handle_exception([](std::exception_ptr e) {

service/migration_task.hh

@@ -47,7 +47,7 @@ namespace service {
 
 class migration_task {
 public:
-    static future<> run_may_throw(const gms::inet_address& endpoint);
+    static future<> run_may_throw(const gms::inet_address& endpoint, bool can_ignore_down_node);
 };
 
 }

service/storage_proxy.cc

@@ -334,6 +334,7 @@ public:
         if (_handler) {
             _handler->prune(_proposal->ballot);
         }
+        _proposal.release();
     }
 };
 
@@ -1661,16 +1662,22 @@ void storage_proxy_stats::stats::register_stats() {
     });
 }
 
-inline uint64_t& storage_proxy_stats::split_stats::get_ep_stat(gms::inet_address ep) {
+inline uint64_t& storage_proxy_stats::split_stats::get_ep_stat(gms::inet_address ep) noexcept {
     if (fbu::is_me(ep)) {
         return _local.val;
     }
 
-    sstring dc = get_dc(ep);
-    if (_auto_register_metrics) {
-        register_metrics_for(ep);
+    try {
+        sstring dc = get_dc(ep);
+        if (_auto_register_metrics) {
+            register_metrics_for(ep);
+        }
+        return _dc_stats[dc].val;
+    } catch (...) {
+        static thread_local uint64_t dummy_stat;
+        slogger.error("Failed to obtain stats ({}), fall-back to dummy", std::current_exception());
+        return dummy_stat;
     }
-    return _dc_stats[dc].val;
 }
 
 void storage_proxy_stats::split_stats::register_metrics_local() {
@@ -5021,14 +5028,15 @@ void storage_proxy::init_messaging_service() {
         }
 
         pruning++;
+        auto d = defer([] { pruning--; });
         return get_schema_for_read(schema_id, src_addr).then([this, key = std::move(key), ballot,
-                         timeout, tr_state = std::move(tr_state), src_ip] (schema_ptr schema) mutable {
+                         timeout, tr_state = std::move(tr_state), src_ip, d = std::move(d)] (schema_ptr schema) mutable {
             dht::token token = dht::get_token(*schema, key);
             unsigned shard = dht::shard_of(*schema, token);
             bool local = shard == engine().cpu_id();
             get_stats().replica_cross_shard_ops += !local;
             return smp::submit_to(shard, _write_smp_service_group, [gs = global_schema_ptr(schema), gt = tracing::global_trace_state_ptr(std::move(tr_state)),
-                                     local,  key = std::move(key), ballot, timeout, src_ip, d = defer([] { pruning--; })] () {
+                                     local,  key = std::move(key), ballot, timeout, src_ip, d = std::move(d)] () {
                 tracing::trace_state_ptr tr_state = gt;
                 return paxos::paxos_state::prune(gs, key, ballot,  *timeout, tr_state).then([src_ip, tr_state] () {
                     tracing::trace(tr_state, "paxos_prune: handling is done, sending a response to /{}", src_ip);
@@ -5042,18 +5050,22 @@ void storage_proxy::init_messaging_service() {
 future<> storage_proxy::uninit_messaging_service() {
     auto& ms = netw::get_local_messaging_service();
     return when_all_succeed(
+        ms.unregister_counter_mutation(),
         ms.unregister_mutation(),
+        ms.unregister_hint_mutation(),
         ms.unregister_mutation_done(),
         ms.unregister_mutation_failed(),
         ms.unregister_read_data(),
         ms.unregister_read_mutation_data(),
         ms.unregister_read_digest(),
         ms.unregister_truncate(),
+        ms.unregister_get_schema_version(),
         ms.unregister_paxos_prepare(),
         ms.unregister_paxos_accept(),
         ms.unregister_paxos_learn(),
         ms.unregister_paxos_prune()
     );
+
 }
 
 future<rpc::tuple<foreign_ptr<lw_shared_ptr<reconcilable_result>>, cache_temperature>>
@@ -5146,8 +5158,7 @@ future<> storage_proxy::drain_on_shutdown() {
 
 future<>
 storage_proxy::stop() {
-    // FIXME: hints manager should be stopped here but it seems like this function is never called
-    return uninit_messaging_service();
+    return make_ready_future<>();
 }
 
 }

service/storage_proxy.hh

@@ -298,7 +298,6 @@ private:
     cdc::cdc_service* _cdc = nullptr;
     cdc_stats _cdc_stats;
 private:
-    future<> uninit_messaging_service();
     future<coordinator_query_result> query_singular(lw_shared_ptr<query::read_command> cmd,
             dht::partition_range_vector&& partition_ranges,
             db::consistency_level cl,
@@ -452,6 +451,7 @@ public:
         return next;
     }
     void init_messaging_service();
+    future<> uninit_messaging_service();
 
     // Applies mutation on this node.
     // Resolves with timed_out_error when timeout is reached.

service/storage_proxy_stats.hh

@@ -74,7 +74,7 @@ public:
      *
      * @return a reference to the requested counter
      */
-    uint64_t& get_ep_stat(gms::inet_address ep);
+    uint64_t& get_ep_stat(gms::inet_address ep) noexcept;
 };
 
 struct write_stats {

service/storage_service.cc

@@ -2175,7 +2175,8 @@ storage_service::get_snapshot_details() {
 }
 
 future<int64_t> storage_service::true_snapshots_size() {
-    return _db.map_reduce(adder<int64_t>(), [] (database& db) {
+  return run_snapshot_list_operation([] {
+    return get_local_storage_service()._db.map_reduce(adder<int64_t>(), [] (database& db) {
         return do_with(int64_t(0), [&db] (auto& local_total) {
             return parallel_for_each(db.get_column_families(), [&local_total] (auto& cf_pair) {
                 return cf_pair.second->get_snapshot_details().then([&local_total] (auto map) {
@@ -2189,6 +2190,7 @@ future<int64_t> storage_service::true_snapshots_size() {
             });
         });
     });
+  });
 }
 
 static std::atomic<bool> isolated = { false };

sstables/compaction.cc

@@ -66,7 +66,6 @@
 #include "db_clock.hh"
 #include "mutation_compactor.hh"
 #include "leveled_manifest.hh"
-#include "utils/observable.hh"
 #include "dht/token.hh"
 
 namespace sstables {
@@ -284,68 +283,73 @@ public:
 // When compaction finishes, all the temporary sstables generated here will be deleted and removed
 // from table's sstable set.
 class garbage_collected_sstable_writer {
-    compaction* _c = nullptr;
-    std::vector<shared_sstable> _temp_sealed_gc_sstables;
-    std::deque<compaction_write_monitor> _active_write_monitors = {};
-    shared_sstable _sst;
-    std::optional<sstable_writer> _writer;
-    std::optional<utils::observer<>> _on_new_sstable_sealed_observer;
-    utils::UUID _run_identifier = utils::make_random_uuid();
-    bool _consuming_new_partition {};
-private:
-    void setup_on_new_sstable_sealed_handler();
-    void maybe_create_new_sstable_writer();
-    void finish_sstable_writer();
-    void on_end_of_stream();
 public:
-    garbage_collected_sstable_writer() = default;
-    explicit garbage_collected_sstable_writer(compaction& c) : _c(&c) {
-        setup_on_new_sstable_sealed_handler();
-    }
+    // Data for GC writer is stored separately to allow compaction class to communicate directly
+    // with garbage_collected_sstable_writer which is moved into mutation_compaction, making it
+    // unreachable after the compaction process has started.
+    class data {
+        compaction* _c = nullptr;
+        // Garbage collected sstables that are sealed but were not added to SSTable set yet.
+        std::vector<shared_sstable> _unused_garbage_collected_sstables;
+        // Garbage collected sstables that were added to SSTable set and should be eventually removed from it.
+        std::vector<shared_sstable> _used_garbage_collected_sstables;
+        std::deque<compaction_write_monitor> _active_write_monitors = {};
+        shared_sstable _sst;
+        std::optional<sstable_writer> _writer;
+        utils::UUID _run_identifier = utils::make_random_uuid();
+    public:
+        explicit data(compaction& c) : _c(&c) {
+        }
+
+        data& operator=(const data&) = delete;
+        data(const data&) = delete;
+
+        void maybe_create_new_sstable_writer();
+        void finish_sstable_writer();
+
+        // Retrieves all unused garbage collected sstables that will be subsequently added
+        // to the SSTable set, and mark them as used.
+        std::vector<shared_sstable> consume_unused_garbage_collected_sstables() {
+            auto unused = std::exchange(_unused_garbage_collected_sstables, {});
+            _used_garbage_collected_sstables.insert(_used_garbage_collected_sstables.end(), unused.begin(), unused.end());
+            return unused;
+        }
+
+        const std::vector<shared_sstable>& used_garbage_collected_sstables() const {
+            return _used_garbage_collected_sstables;
+        }
+
+        friend class garbage_collected_sstable_writer;
+    };
+private:
+    garbage_collected_sstable_writer::data* _data = nullptr;
+public:
+    explicit garbage_collected_sstable_writer() = default;
+    explicit garbage_collected_sstable_writer(garbage_collected_sstable_writer::data& data) : _data(&data) {}
 
     garbage_collected_sstable_writer& operator=(const garbage_collected_sstable_writer&) = delete;
     garbage_collected_sstable_writer(const garbage_collected_sstable_writer&) = delete;
 
-    garbage_collected_sstable_writer(garbage_collected_sstable_writer&& other)
-            : _c(other._c)
-            , _temp_sealed_gc_sstables(std::move(other._temp_sealed_gc_sstables))
-            , _active_write_monitors(std::move(other._active_write_monitors))
-            , _sst(std::move(other._sst))
-            , _writer(std::move(other._writer))
-            , _run_identifier(other._run_identifier)
-            , _consuming_new_partition(other._consuming_new_partition) {
-        other._on_new_sstable_sealed_observer->disconnect();
-        setup_on_new_sstable_sealed_handler();
-    }
-
-    garbage_collected_sstable_writer& operator=(garbage_collected_sstable_writer&& other) {
-        if (this != &other) {
-            this->~garbage_collected_sstable_writer();
-            new (this) garbage_collected_sstable_writer(std::move(other));
-        }
-        return *this;
-    }
+    garbage_collected_sstable_writer(garbage_collected_sstable_writer&& other) = default;
+    garbage_collected_sstable_writer& operator=(garbage_collected_sstable_writer&& other) = default;
 
     void consume_new_partition(const dht::decorated_key& dk) {
-        maybe_create_new_sstable_writer();
-        _writer->consume_new_partition(dk);
-        _consuming_new_partition = true;
+        _data->maybe_create_new_sstable_writer();
+        _data->_writer->consume_new_partition(dk);
     }
 
-    void consume(tombstone t) { _writer->consume(t); }
-    stop_iteration consume(static_row&& sr, tombstone, bool) { return _writer->consume(std::move(sr)); }
-    stop_iteration consume(clustering_row&& cr, row_tombstone, bool) { return _writer->consume(std::move(cr)); }
-    stop_iteration consume(range_tombstone&& rt) { return _writer->consume(std::move(rt)); }
+    void consume(tombstone t) { _data->_writer->consume(t); }
+    stop_iteration consume(static_row&& sr, tombstone, bool) { return _data->_writer->consume(std::move(sr)); }
+    stop_iteration consume(clustering_row&& cr, row_tombstone, bool) { return _data->_writer->consume(std::move(cr)); }
+    stop_iteration consume(range_tombstone&& rt) { return _data->_writer->consume(std::move(rt)); }
 
     stop_iteration consume_end_of_partition() {
-        _writer->consume_end_of_partition();
-        _consuming_new_partition = false;
+        _data->_writer->consume_end_of_partition();
         return stop_iteration::no;
     }
 
     void consume_end_of_stream() {
-        finish_sstable_writer();
-        on_end_of_stream();
+        _data->finish_sstable_writer();
     }
 };
 
@@ -380,6 +384,7 @@ public:
 class compaction {
 protected:
     column_family& _cf;
+    creator_fn _sstable_creator;
     schema_ptr _schema;
     std::vector<shared_sstable> _sstables;
     // Unused sstables are tracked because if compaction is interrupted we can only delete them.
@@ -393,15 +398,17 @@ protected:
     std::vector<unsigned long> _ancestors;
     db::replay_position _rp;
     encoding_stats_collector _stats_collector;
-    utils::observable<> _on_new_sstable_sealed;
     bool _contains_multi_fragment_runs = false;
+    garbage_collected_sstable_writer::data _gc_sstable_writer_data;
 protected:
-    compaction(column_family& cf, std::vector<shared_sstable> sstables, uint64_t max_sstable_size, uint32_t sstable_level)
+    compaction(column_family& cf, creator_fn creator, std::vector<shared_sstable> sstables, uint64_t max_sstable_size, uint32_t sstable_level)
         : _cf(cf)
+        , _sstable_creator(std::move(creator))
         , _schema(cf.schema())
         , _sstables(std::move(sstables))
         , _max_sstable_size(max_sstable_size)
         , _sstable_level(sstable_level)
+        , _gc_sstable_writer_data(*this)
     {
         _info->cf = &cf;
         for (auto& sst : _sstables) {
@@ -434,10 +441,6 @@ protected:
         writer = std::nullopt;
         sst->open_data().get0();
         _info->end_size += sst->bytes_on_disk();
-        // Notify GC'ed-data sstable writer's handler that an output sstable has just been sealed.
-        // The handler is responsible for making sure that deleting an input sstable will not
-        // result in resurrection on failure.
-        _on_new_sstable_sealed();
     }
 
     api::timestamp_type maximum_timestamp() const {
@@ -447,10 +450,6 @@ protected:
         return (*m)->get_stats_metadata().max_timestamp;
     }
 
-    utils::observer<> add_on_new_sstable_sealed_handler(std::function<void (void)> handler) noexcept {
-        return _on_new_sstable_sealed.observe(std::move(handler));
-    }
-
     encoding_stats get_encoding_stats() const {
         return _stats_collector.get();
     }
@@ -562,10 +561,9 @@ private:
         };
     }
 
-    virtual shared_sstable create_new_sstable() const = 0;
-
     // select a sstable writer based on decorated key.
     virtual sstable_writer* select_sstable_writer(const dht::decorated_key& dk) = 0;
+
     // stop current writer
     virtual void stop_sstable_writer() = 0;
     // finish all writers.
@@ -588,20 +586,9 @@ private:
             sst->mark_for_deletion();
         }
     }
-
-    void setup_garbage_collected_sstable(shared_sstable sst) {
-        // Add new sstable to table's set because expired tombstone should be available if compaction is abruptly stopped.
-        _cf.add_sstable_and_update_cache(std::move(sst)).get();
-    }
-
-    void eventually_delete_garbage_collected_sstable(shared_sstable sst) {
-        // Add sstable to compaction's input list for it to be eventually removed from table's set.
-        sst->mark_for_deletion();
-        _sstables.push_back(std::move(sst));
-    }
 public:
     garbage_collected_sstable_writer make_garbage_collected_sstable_writer() {
-        return garbage_collected_sstable_writer(*this);
+        return garbage_collected_sstable_writer(_gc_sstable_writer_data);
     }
 
     bool contains_multi_fragment_runs() const {
@@ -616,6 +603,7 @@ public:
 
     friend class compacting_sstable_writer;
     friend class garbage_collected_sstable_writer;
+    friend class garbage_collected_sstable_writer::data;
 };
 
 void compacting_sstable_writer::consume_new_partition(const dht::decorated_key& dk) {
@@ -642,22 +630,9 @@ void compacting_sstable_writer::consume_end_of_stream() {
     _c.finish_sstable_writer();
 }
 
-void garbage_collected_sstable_writer::setup_on_new_sstable_sealed_handler() {
-    _on_new_sstable_sealed_observer = _c->add_on_new_sstable_sealed_handler([this] {
-        // NOTE: This handler is called, BEFORE an input sstable is possibly deleted
-        // *AND* AFTER a new output sstable is sealed, to flush a garbage collected
-        // sstable being currently written.
-        // That way, data is resurrection is prevented by making sure that the
-        // GC'able data is still reachable in a temporary sstable.
-        assert(!_consuming_new_partition);
-        // Wait for current gc'ed-only-sstable to be flushed and added to table's set.
-        this->finish_sstable_writer();
-    });
-}
-
-void garbage_collected_sstable_writer::maybe_create_new_sstable_writer() {
+void garbage_collected_sstable_writer::data::maybe_create_new_sstable_writer() {
     if (!_writer) {
-        _sst = _c->create_new_sstable();
+        _sst = _c->_sstable_creator(this_shard_id());
 
         auto&& priority = service::get_local_compaction_priority();
         _active_write_monitors.emplace_back(_sst, _c->_cf, _c->maximum_timestamp(), _c->_sstable_level);
@@ -668,25 +643,16 @@ void garbage_collected_sstable_writer::maybe_create_new_sstable_writer() {
     }
 }
 
-void garbage_collected_sstable_writer::finish_sstable_writer() {
+void garbage_collected_sstable_writer::data::finish_sstable_writer() {
     if (_writer) {
         _writer->consume_end_of_stream();
         _writer = std::nullopt;
         _sst->open_data().get0();
-        _c->setup_garbage_collected_sstable(_sst);
-        _temp_sealed_gc_sstables.push_back(std::move(_sst));
-    }
-}
-
-void garbage_collected_sstable_writer::on_end_of_stream() {
-    for (auto&& sst : _temp_sealed_gc_sstables) {
-        clogger.debug("Asking for deletion of temporary tombstone-only sstable {}", sst->get_filename());
-        _c->eventually_delete_garbage_collected_sstable(std::move(sst));
+        _unused_garbage_collected_sstables.push_back(std::move(_sst));
     }
 }
 
 class regular_compaction : public compaction {
-    std::function<shared_sstable()> _creator;
     replacer_fn _replacer;
     std::unordered_set<shared_sstable> _compacting_for_max_purgeable_func;
     // store a clone of sstable set for column family, which needs to be alive for incremental selector.
@@ -701,10 +667,9 @@ class regular_compaction : public compaction {
     std::deque<compaction_write_monitor> _active_write_monitors = {};
     utils::UUID _run_identifier;
 public:
-    regular_compaction(column_family& cf, compaction_descriptor descriptor, std::function<shared_sstable()> creator, replacer_fn replacer)
-        : compaction(cf, std::move(descriptor.sstables), descriptor.max_sstable_bytes, descriptor.level)
-        , _creator(std::move(creator))
-        , _replacer(std::move(replacer))
+    regular_compaction(column_family& cf, compaction_descriptor descriptor)
+        : compaction(cf, std::move(descriptor.creator), std::move(descriptor.sstables), descriptor.max_sstable_bytes, descriptor.level)
+        , _replacer(std::move(descriptor.replacer))
         , _compacting_for_max_purgeable_func(std::unordered_set<shared_sstable>(_sstables.begin(), _sstables.end()))
         , _set(cf.get_sstable_set())
         , _selector(_set.make_incremental_selector())
@@ -755,13 +720,9 @@ public:
         };
     }
 
-    virtual shared_sstable create_new_sstable() const override {
-        return _creator();
-    }
-
     virtual sstable_writer* select_sstable_writer(const dht::decorated_key& dk) override {
         if (!_writer) {
-            _sst = _creator();
+            _sst = _sstable_creator(0);
             setup_new_sstable(_sst);
 
             _active_write_monitors.emplace_back(_sst, _cf, maximum_timestamp(), _sstable_level);
@@ -834,6 +795,15 @@ private:
                 // Fully expired sstable is not actually compacted, therefore it's not present in the compacting set.
                 _compacting->erase(sst);
             });
+            // Make sure SSTable created by garbage collected writer is made available
+            // before exhausted SSTable is released, so as to prevent data resurrection.
+            _gc_sstable_writer_data.finish_sstable_writer();
+            // Added Garbage collected SSTables to list of unused SSTables that will be added
+            // to SSTable set. GC SSTables should be added before compaction completes because
+            // a failure could result in data resurrection if data is not made available.
+            auto unused_gc_sstables = _gc_sstable_writer_data.consume_unused_garbage_collected_sstables();
+            _new_unused_sstables.insert(_new_unused_sstables.end(), unused_gc_sstables.begin(), unused_gc_sstables.end());
+
             auto exhausted_ssts = std::vector<shared_sstable>(exhausted, _sstables.end());
             _replacer(get_compaction_completion_desc(exhausted_ssts, std::move(_new_unused_sstables)));
             _sstables.erase(exhausted, _sstables.end());
@@ -842,11 +812,16 @@ private:
     }
 
     void replace_remaining_exhausted_sstables() {
-        if (!_sstables.empty()) {
-            std::vector<shared_sstable> sstables_compacted;
-            std::move(_sstables.begin(), _sstables.end(), std::back_inserter(sstables_compacted));
-            _replacer(get_compaction_completion_desc(std::move(sstables_compacted), std::move(_new_unused_sstables)));
-        }
+        if (!_sstables.empty() || !_gc_sstable_writer_data.used_garbage_collected_sstables().empty()) {
+            std::vector<shared_sstable> old_sstables;
+            std::move(_sstables.begin(), _sstables.end(), std::back_inserter(old_sstables));
+
+            // Remove Garbage Collected SSTables from the SSTable set if any was previously added.
+            auto& used_garbage_collected_sstables = _gc_sstable_writer_data.used_garbage_collected_sstables();
+            old_sstables.insert(old_sstables.end(), used_garbage_collected_sstables.begin(), used_garbage_collected_sstables.end());
+
+            _replacer(get_compaction_completion_desc(std::move(old_sstables), std::move(_new_unused_sstables)));
+         }
     }
 
     void do_pending_replacements() {
@@ -909,8 +884,8 @@ protected:
         return compaction_completion_desc{std::move(input_sstables), std::move(output_sstables), std::move(ranges_for_for_invalidation)};
     }
 public:
-    cleanup_compaction(column_family& cf, compaction_descriptor descriptor, std::function<shared_sstable()> creator, replacer_fn replacer)
-        : regular_compaction(cf, std::move(descriptor), std::move(creator), std::move(replacer))
+    cleanup_compaction(column_family& cf, compaction_descriptor descriptor)
+        : regular_compaction(cf, std::move(descriptor))
         , _owned_ranges(service::get_local_storage_service().get_local_ranges(_schema->ks_name()))
     {
         _info->type = compaction_type::Cleanup;
@@ -1114,9 +1089,8 @@ private:
     compaction_options::scrub _options;
 
 public:
-    scrub_compaction(column_family& cf, compaction_descriptor descriptor, compaction_options::scrub options, std::function<shared_sstable()> creator,
-            replacer_fn replacer)
-        : regular_compaction(cf, std::move(descriptor), std::move(creator), std::move(replacer))
+    scrub_compaction(column_family& cf, compaction_descriptor descriptor, compaction_options::scrub options)
+        : regular_compaction(cf, std::move(descriptor))
         , _options(options) {
         _info->type = compaction_type::Scrub;
     }
@@ -1143,7 +1117,6 @@ flat_mutation_reader make_scrubbing_reader(flat_mutation_reader rd, bool skip_co
 class resharding_compaction final : public compaction {
     std::vector<std::pair<shared_sstable, std::optional<sstable_writer>>> _output_sstables;
     shard_id _shard; // shard of current sstable writer
-    std::function<shared_sstable(shard_id)> _sstable_creator;
     compaction_backlog_tracker _resharding_backlog_tracker;
 
     // Partition count estimation for a shard S:
@@ -1168,11 +1141,9 @@ private:
         return ceil(double(_estimation_per_shard[s].estimated_partitions) / estimated_sstables);
     }
 public:
-    resharding_compaction(std::vector<shared_sstable> sstables, column_family& cf, std::function<shared_sstable(shard_id)> creator,
-            uint64_t max_sstable_size, uint32_t sstable_level)
-        : compaction(cf, std::move(sstables), max_sstable_size, sstable_level)
+    resharding_compaction(column_family& cf, sstables::compaction_descriptor descriptor)
+        : compaction(cf, std::move(descriptor.creator), std::move(descriptor.sstables), descriptor.max_sstable_bytes, descriptor.level)
         , _output_sstables(smp::count)
-        , _sstable_creator(std::move(creator))
         , _resharding_backlog_tracker(std::make_unique<resharding_backlog_tracker>())
         , _estimation_per_shard(smp::count)
         , _run_identifiers(smp::count)
@@ -1224,10 +1195,6 @@ public:
 
     void backlog_tracker_adjust_charges() override { }
 
-    shared_sstable create_new_sstable() const override {
-        return _sstable_creator(_shard);
-    }
-
     sstable_writer* select_sstable_writer(const dht::decorated_key& dk) override {
         _shard = dht::shard_of(*_schema, dk.token());
         auto& sst = _output_sstables[_shard].first;
@@ -1298,38 +1265,35 @@ compaction_type compaction_options::type() const {
     return index_to_type[_options.index()];
 }
 
-static std::unique_ptr<compaction> make_compaction(column_family& cf, sstables::compaction_descriptor descriptor,
-        std::function<shared_sstable()> creator, replacer_fn replacer) {
+static std::unique_ptr<compaction> make_compaction(column_family& cf, sstables::compaction_descriptor descriptor) {
     struct {
         column_family& cf;
         sstables::compaction_descriptor&& descriptor;
-        std::function<shared_sstable()>&& creator;
-        replacer_fn&& replacer;
 
         std::unique_ptr<compaction> operator()(compaction_options::regular) {
-            return std::make_unique<regular_compaction>(cf, std::move(descriptor), std::move(creator), std::move(replacer));
+            return std::make_unique<regular_compaction>(cf, std::move(descriptor));
         }
         std::unique_ptr<compaction> operator()(compaction_options::cleanup) {
-            return std::make_unique<cleanup_compaction>(cf, std::move(descriptor), std::move(creator), std::move(replacer));
+            return std::make_unique<cleanup_compaction>(cf, std::move(descriptor));
         }
         std::unique_ptr<compaction> operator()(compaction_options::upgrade) {
-            return std::make_unique<cleanup_compaction>(cf, std::move(descriptor), std::move(creator), std::move(replacer));
+            return std::make_unique<cleanup_compaction>(cf, std::move(descriptor));
         }
         std::unique_ptr<compaction> operator()(compaction_options::scrub scrub_options) {
-            return std::make_unique<scrub_compaction>(cf, std::move(descriptor), scrub_options, std::move(creator), std::move(replacer));
+            return std::make_unique<scrub_compaction>(cf, std::move(descriptor), scrub_options);
         }
-    } visitor_factory{cf, std::move(descriptor), std::move(creator), std::move(replacer)};
+    } visitor_factory{cf, std::move(descriptor)};
 
     return descriptor.options.visit(visitor_factory);
 }
 
 future<compaction_info>
-compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf, std::function<shared_sstable()> creator, replacer_fn replacer) {
+compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf) {
     if (descriptor.sstables.empty()) {
         throw std::runtime_error(format("Called {} compaction with empty set on behalf of {}.{}", compaction_name(descriptor.options.type()),
                 cf.schema()->ks_name(), cf.schema()->cf_name()));
     }
-    auto c = make_compaction(cf, std::move(descriptor), std::move(creator), std::move(replacer));
+    auto c = make_compaction(cf, std::move(descriptor));
     if (c->contains_multi_fragment_runs()) {
         auto gc_writer = c->make_garbage_collected_sstable_writer();
         return compaction::run(std::move(c), std::move(gc_writer));
@@ -1343,7 +1307,10 @@ reshard_sstables(std::vector<shared_sstable> sstables, column_family& cf, std::f
     if (sstables.empty()) {
         throw std::runtime_error(format("Called resharding with empty set on behalf of {}.{}", cf.schema()->ks_name(), cf.schema()->cf_name()));
     }
-    auto c = std::make_unique<resharding_compaction>(std::move(sstables), cf, std::move(creator), max_sstable_size, sstable_level);
+    sstables::compaction_descriptor descriptor(std::move(sstables), sstable_level, max_sstable_size);
+    descriptor.creator = std::move(creator);
+
+    auto c = std::make_unique<resharding_compaction>(cf, std::move(descriptor));
     return compaction::run(std::move(c)).then([] (auto ret) {
         return std::move(ret.new_sstables);
     });

sstables/compaction.hh

@@ -84,6 +84,20 @@ namespace sstables {
         compaction_type type() const;
     };
 
+    struct compaction_completion_desc {
+        // Old, existing SSTables that should be deleted and removed from the SSTable set.
+        std::vector<shared_sstable> old_sstables;
+        // New, fresh SSTables that should be added to SSTable set, replacing the old ones.
+        std::vector<shared_sstable> new_sstables;
+        // Set of compacted partition ranges that should be invalidated in the cache.
+        dht::partition_range_vector ranges_for_cache_invalidation;
+    };
+
+    // creates a new SSTable for a given shard
+    using creator_fn = std::function<shared_sstable(shard_id shard)>;
+    // Replaces old sstable(s) by new one(s) which contain all non-expired data.
+    using replacer_fn = std::function<void(compaction_completion_desc)>;
+
     struct compaction_descriptor {
         // List of sstables to be compacted.
         std::vector<sstables::shared_sstable> sstables;
@@ -101,6 +115,9 @@ namespace sstables {
         // This also selects the kind of compaction to do.
         compaction_options options = compaction_options::make_regular();
 
+        creator_fn creator;
+        replacer_fn replacer;
+
         compaction_descriptor() = default;
 
         static constexpr int default_level = 0;
@@ -190,16 +207,6 @@ namespace sstables {
         }
     };
 
-    struct compaction_completion_desc {
-        std::vector<shared_sstable> input_sstables;
-        std::vector<shared_sstable> output_sstables;
-        // Set of compacted partition ranges that should be invalidated in the cache.
-        dht::partition_range_vector ranges_for_cache_invalidation;
-    };
-
-    // Replaces old sstable(s) by new one(s) which contain all non-expired data.
-    using replacer_fn = std::function<void(compaction_completion_desc)>;
-
     // Compact a list of N sstables into M sstables.
     // Returns info about the finished compaction, which includes vector to new sstables.
     //
@@ -212,8 +219,7 @@ namespace sstables {
     // If descriptor.cleanup is true, mutation that doesn't belong to current node will be
     // cleaned up, log messages will inform the user that compact_sstables runs for
     // cleaning operation, and compaction history will not be updated.
-    future<compaction_info> compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf,
-        std::function<shared_sstable()> creator, replacer_fn replacer);
+    future<compaction_info> compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf);
 
     // Compacts a set of N shared sstables into M sstables. For every shard involved,
     // i.e. which owns any of the sstables, a new unshared sstable is created.

sstables/compaction_backlog_manager.hh

@@ -92,6 +92,9 @@ public:
     void transfer_ongoing_charges(compaction_backlog_tracker& new_bt, bool move_read_charges = true);
     void revert_charges(sstables::shared_sstable sst);
 private:
+    // Returns true if this SSTable can be added or removed from the tracker.
+    bool sstable_belongs_to_tracker(const sstables::shared_sstable& sst);
+
     void disable() {
         _disabled = true;
         _ongoing_writes = {};

sstables/compaction_manager.cc

@@ -218,7 +218,7 @@ std::vector<sstables::shared_sstable> compaction_manager::get_candidates(const c
     auto& cs = cf.get_compaction_strategy();
 
     // Filter out sstables that are being compacted.
-    for (auto& sst : cf.candidates_for_compaction()) {
+    for (auto& sst : cf.non_staging_sstables()) {
         if (_compacting_sstables.count(sst)) {
             continue;
         }
@@ -646,8 +646,8 @@ future<> compaction_manager::rewrite_sstables(column_family* cf, sstables::compa
     return task->compaction_done.get_future().then([task] {});
 }
 
-static bool needs_cleanup(const sstables::shared_sstable& sst,
-                   const dht::token_range_vector& owned_ranges,
+bool needs_cleanup(const sstables::shared_sstable& sst,
+                   const dht::token_range_vector& sorted_owned_ranges,
                    schema_ptr s) {
     auto first = sst->get_first_partition_key();
     auto last = sst->get_last_partition_key();
@@ -655,29 +655,40 @@ static bool needs_cleanup(const sstables::shared_sstable& sst,
     auto last_token = dht::get_token(*s, last);
     dht::token_range sst_token_range = dht::token_range::make(first_token, last_token);
 
+    auto r = std::lower_bound(sorted_owned_ranges.begin(), sorted_owned_ranges.end(), first_token,
+            [] (const range<dht::token>& a, const dht::token& b) {
+        // check that range a is before token b.
+        return a.after(b, dht::token_comparator());
+    });
+
     // return true iff sst partition range isn't fully contained in any of the owned ranges.
-    for (auto& r : owned_ranges) {
-        if (r.contains(sst_token_range, dht::token_comparator())) {
+    if (r != sorted_owned_ranges.end()) {
+        if (r->contains(sst_token_range, dht::token_comparator())) {
             return false;
         }
     }
     return true;
 }
 
-future<> compaction_manager::perform_cleanup(column_family* cf) {
+future<> compaction_manager::perform_cleanup(database& db, column_family* cf) {
     if (check_for_cleanup(cf)) {
         throw std::runtime_error(format("cleanup request failed: there is an ongoing cleanup on {}.{}",
             cf->schema()->ks_name(), cf->schema()->cf_name()));
     }
-    return rewrite_sstables(cf, sstables::compaction_options::make_cleanup(), [this] (const table& table) {
-        auto schema = table.schema();
-        auto owned_ranges = service::get_local_storage_service().get_local_ranges(schema->ks_name());
+    return seastar::async([this, cf, &db] {
+        auto schema = cf->schema();
+        auto& rs = db.find_keyspace(schema->ks_name()).get_replication_strategy();
+        auto sorted_owned_ranges = rs.get_ranges_in_thread(utils::fb_utilities::get_broadcast_address());
         auto sstables = std::vector<sstables::shared_sstable>{};
-        const auto candidates = table.candidates_for_compaction();
-        std::copy_if(candidates.begin(), candidates.end(), std::back_inserter(sstables), [&owned_ranges, schema] (const sstables::shared_sstable& sst) {
-            return owned_ranges.empty() || needs_cleanup(sst, owned_ranges, schema);
+        const auto candidates = get_candidates(*cf);
+        std::copy_if(candidates.begin(), candidates.end(), std::back_inserter(sstables), [&sorted_owned_ranges, schema] (const sstables::shared_sstable& sst) {
+            seastar::thread::maybe_yield();
+            return sorted_owned_ranges.empty() || needs_cleanup(sst, sorted_owned_ranges, schema);
         });
         return sstables;
+    }).then([this, cf] (std::vector<sstables::shared_sstable> sstables) {
+        return rewrite_sstables(cf, sstables::compaction_options::make_cleanup(),
+                [sstables = std::move(sstables)] (const table&) { return sstables; });
     });
 }
 
@@ -692,7 +703,7 @@ future<> compaction_manager::perform_sstable_upgrade(column_family* cf, bool exc
         return cf->run_with_compaction_disabled([this, cf, &tables, exclude_current_version] {
             auto last_version = cf->get_sstables_manager().get_highest_supported_format();
 
-            for (auto& sst : cf->candidates_for_compaction()) {
+            for (auto& sst : get_candidates(*cf)) {
                 // if we are a "normal" upgrade, we only care about
                 // tables with older versions, but potentially
                 // we are to actually rewrite everything. (-a)
@@ -708,8 +719,8 @@ future<> compaction_manager::perform_sstable_upgrade(column_family* cf, bool exc
             // Note that we potentially could be doing multiple
             // upgrades here in parallel, but that is really the users
             // problem.
-            return rewrite_sstables(cf, sstables::compaction_options::make_upgrade(), [&](auto&) {
-                return tables;
+            return rewrite_sstables(cf, sstables::compaction_options::make_upgrade(), [&](auto&) mutable {
+                return std::exchange(tables, {});
             });
         });
     });
@@ -717,8 +728,8 @@ future<> compaction_manager::perform_sstable_upgrade(column_family* cf, bool exc
 
 // Submit a column family to be scrubbed and wait for its termination.
 future<> compaction_manager::perform_sstable_scrub(column_family* cf, bool skip_corrupted) {
-    return rewrite_sstables(cf, sstables::compaction_options::make_scrub(skip_corrupted), [] (const table& cf) {
-        return cf.candidates_for_compaction();
+    return rewrite_sstables(cf, sstables::compaction_options::make_scrub(skip_corrupted), [this] (const table& cf) {
+        return get_candidates(cf);
     });
 }
 
@@ -795,7 +806,7 @@ double compaction_backlog_tracker::backlog() const {
 }
 
 void compaction_backlog_tracker::add_sstable(sstables::shared_sstable sst) {
-    if (_disabled) {
+    if (_disabled || !sstable_belongs_to_tracker(sst)) {
         return;
     }
     _ongoing_writes.erase(sst);
@@ -808,7 +819,7 @@ void compaction_backlog_tracker::add_sstable(sstables::shared_sstable sst) {
 }
 
 void compaction_backlog_tracker::remove_sstable(sstables::shared_sstable sst) {
-    if (_disabled) {
+    if (_disabled || !sstable_belongs_to_tracker(sst)) {
         return;
     }
 
@@ -821,6 +832,10 @@ void compaction_backlog_tracker::remove_sstable(sstables::shared_sstable sst) {
     }
 }
 
+bool compaction_backlog_tracker::sstable_belongs_to_tracker(const sstables::shared_sstable& sst) {
+    return !sst->requires_view_building();
+}
+
 void compaction_backlog_tracker::register_partially_written_sstable(sstables::shared_sstable sst, backlog_write_progress_manager& wp) {
     if (_disabled) {
         return;

sstables/compaction_manager.hh

@@ -175,7 +175,7 @@ public:
     // Cleanup is about discarding keys that are no longer relevant for a
     // given sstable, e.g. after node loses part of its token range because
     // of a newly added node.
-    future<> perform_cleanup(column_family* cf);
+    future<> perform_cleanup(database& db, column_family* cf);
 
     // Submit a column family to be upgraded and wait for its termination.
     future<> perform_sstable_upgrade(column_family* cf, bool exclude_current_version);
@@ -243,3 +243,5 @@ public:
     friend class compaction_weight_registration;
 };
 
+bool needs_cleanup(const sstables::shared_sstable& sst, const dht::token_range_vector& owned_ranges, schema_ptr s);
+

sstables/compaction_strategy.cc

@@ -792,7 +792,11 @@ uint64_t time_window_compaction_strategy::adjust_partition_estimate(const mutati
     }
     const auto min_window = get_window_for(_options, *ms_meta.min_timestamp);
     const auto max_window = get_window_for(_options, *ms_meta.max_timestamp);
-    return partition_estimate / (max_window - min_window + 1);
+    const auto window_size = get_window_size(_options);
+
+    auto estimated_window_count = (max_window + (window_size - 1) - min_window) / window_size;
+
+    return partition_estimate / std::max(1UL, uint64_t(estimated_window_count));
 }
 
 namespace {

sstables/index_reader.hh

@@ -401,9 +401,16 @@ private:
                     auto indexes = std::move(entries_reader->_consumer.indexes);
                     return entries_reader->_context.close().then([indexes = std::move(indexes), ex = std::move(ex)] () mutable {
                         if (ex) {
-                            std::rethrow_exception(std::move(ex));
+                            return do_with(std::move(indexes), [ex = std::move(ex)] (index_list& indexes) mutable {
+                                return parallel_for_each(indexes, [] (index_entry& ie) mutable {
+                                    return ie.close_pi_stream();
+                                }).then_wrapped([ex = std::move(ex)] (future<>&& fut) mutable {
+                                    fut.ignore_ready_future();
+                                    return make_exception_future<index_list>(std::move(ex));
+                                });
+                            });
                         }
-                        return std::move(indexes);
+                        return make_ready_future<index_list>(std::move(indexes));
                     });
 
                 });

sstables/partition.cc

@@ -67,9 +67,13 @@ data_consume_rows<data_consume_rows_context_m>(const schema& s, shared_sstable,
 static
 position_in_partition_view get_slice_upper_bound(const schema& s, const query::partition_slice& slice, dht::ring_position_view key) {
     const auto& ranges = slice.row_ranges(s, *key.key());
-    return ranges.empty()
-        ? position_in_partition_view::for_static_row()
-        : position_in_partition_view::for_range_end(ranges.back());
+    if (ranges.empty()) {
+        return position_in_partition_view::for_static_row();
+    }
+    if (slice.options.contains(query::partition_slice::option::reversed)) {
+        return position_in_partition_view::for_range_end(ranges.front());
+    }
+    return position_in_partition_view::for_range_end(ranges.back());
 }
 
 GCC6_CONCEPT(

sstables/time_window_compaction_strategy.hh

@@ -305,6 +305,11 @@ public:
     get_window_for(const time_window_compaction_strategy_options& options, api::timestamp_type ts) {
         return get_window_lower_bound(options.sstable_window_size, to_timestamp_type(options.timestamp_resolution, ts));
     }
+
+    static api::timestamp_type
+    get_window_size(const time_window_compaction_strategy_options& options) {
+        return timestamp_type(std::chrono::duration_cast<std::chrono::microseconds>(options.get_sstable_window_size()).count());
+    }
 private:
     void update_estimated_compaction_by_tasks(std::map<timestamp_type, std::vector<shared_sstable>>& tasks, int min_threshold) {
         int64_t n = 0;

streaming/stream_session.cc

@@ -319,6 +319,15 @@ void stream_session::init_messaging_service_handler() {
     });
 }
 
+future<> stream_session::uninit_messaging_service_handler() {
+    return when_all_succeed(
+        ms().unregister_prepare_message(),
+        ms().unregister_prepare_done_message(),
+        ms().unregister_stream_mutation_fragments(),
+        ms().unregister_stream_mutation_done(),
+        ms().unregister_complete_message()).discard_result();
+}
+
 distributed<database>* stream_session::_db;
 distributed<db::system_distributed_keyspace>* stream_session::_sys_dist_ks;
 distributed<db::view::view_update_generator>* stream_session::_view_update_generator;
@@ -342,9 +351,13 @@ future<> stream_session::init_streaming_service(distributed<database>& db, distr
     // });
     return get_stream_manager().start().then([] {
         gms::get_local_gossiper().register_(get_local_stream_manager().shared_from_this());
-        return _db->invoke_on_all([] (auto& db) {
-            init_messaging_service_handler();
-        });
+        return smp::invoke_on_all([] { init_messaging_service_handler(); });
+    });
+}
+
+future<> stream_session::uninit_streaming_service() {
+    return smp::invoke_on_all([] {
+        return uninit_messaging_service_handler();
     });
 }
 

streaming/stream_session.hh

@@ -142,6 +142,7 @@ private:
     using token = dht::token;
     using ring_position = dht::ring_position;
     static void init_messaging_service_handler();
+    static future<> uninit_messaging_service_handler();
     static distributed<database>* _db;
     static distributed<db::system_distributed_keyspace>* _sys_dist_ks;
     static distributed<db::view::view_update_generator>* _view_update_generator;
@@ -152,6 +153,7 @@ public:
     static database& get_local_db() { return _db->local(); }
     static distributed<database>& get_db() { return *_db; };
     static future<> init_streaming_service(distributed<database>& db, distributed<db::system_distributed_keyspace>& sys_dist_ks, distributed<db::view::view_update_generator>& view_update_generator);
+    static future<> uninit_streaming_service();
 public:
     /**
      * Streaming endpoint.

streaming/stream_transfer_task.cc

@@ -44,6 +44,7 @@
 #include "streaming/stream_reason.hh"
 #include "streaming/stream_mutation_fragments_cmd.hh"
 #include "mutation_reader.hh"
+#include "flat_mutation_reader.hh"
 #include "frozen_mutation.hh"
 #include "mutation.hh"
 #include "message/messaging_service.hh"
@@ -203,15 +204,27 @@ future<> send_mutation_fragments(lw_shared_ptr<send_info> si) {
         }();
 
         auto sink_op = [sink, si, got_error_from_peer] () mutable -> future<> {
-            return do_with(std::move(sink), [si, got_error_from_peer] (rpc::sink<frozen_mutation_fragment, stream_mutation_fragments_cmd>& sink) {
-                return repeat([&sink, si, got_error_from_peer] () mutable {
-                    return si->reader(db::no_timeout).then([&sink, si, s = si->reader.schema(), got_error_from_peer] (mutation_fragment_opt mf) mutable {
-                        if (mf && !(*got_error_from_peer)) {
+            mutation_fragment_stream_validator validator(*(si->reader.schema()));
+            return do_with(std::move(sink), std::move(validator), [si, got_error_from_peer] (rpc::sink<frozen_mutation_fragment, stream_mutation_fragments_cmd>& sink, mutation_fragment_stream_validator& validator) {
+                return repeat([&sink, &validator, si, got_error_from_peer] () mutable {
+                    return si->reader(db::no_timeout).then([&sink, &validator, si, s = si->reader.schema(), got_error_from_peer] (mutation_fragment_opt mf) mutable {
+                        if (*got_error_from_peer) {
+                            return make_exception_future<stop_iteration>(std::runtime_error("Got status error code from peer"));
+                        }
+                        if (mf) {
+                            if (!validator(mf->mutation_fragment_kind())) {
+                                return make_exception_future<stop_iteration>(std::runtime_error(format("Stream reader mutation_fragment validator failed, previous={}, current={}",
+                                        validator.previous_mutation_fragment_kind(), mf->mutation_fragment_kind())));
+                            }
                             frozen_mutation_fragment fmf = freeze(*s, *mf);
                             auto size = fmf.representation().size();
                             streaming::get_local_stream_manager().update_progress(si->plan_id, si->id.addr, streaming::progress_info::direction::OUT, size);
                             return sink(fmf, stream_mutation_fragments_cmd::mutation_fragment_data).then([] { return stop_iteration::no; });
                         } else {
+                            if (!validator.on_end_of_stream()) {
+                                return make_exception_future<stop_iteration>(std::runtime_error(format("Stream reader mutation_fragment validator failed on end_of_stream, previous={}, current=end_of_stream",
+                                        validator.previous_mutation_fragment_kind())));
+                            }
                             return make_ready_future<stop_iteration>(stop_iteration::yes);
                         }
                     });

table.cc

@@ -1193,7 +1193,7 @@ table::on_compaction_completion(sstables::compaction_completion_desc& desc) {
     // unbounded time, because all shards must agree on the deletion).
 
     // make sure all old sstables belong *ONLY* to current shard before we proceed to their deletion.
-    for (auto& sst : desc.input_sstables) {
+    for (auto& sst : desc.old_sstables) {
         auto shards = sst->get_shards_for_this_sstable();
         if (shards.size() > 1) {
             throw std::runtime_error(format("A regular compaction for {}.{} INCORRECTLY used shared sstable {}. Only resharding work with those!",
@@ -1207,11 +1207,11 @@ table::on_compaction_completion(sstables::compaction_completion_desc& desc) {
 
     auto new_compacted_but_not_deleted = _sstables_compacted_but_not_deleted;
     // rebuilding _sstables_compacted_but_not_deleted first to make the entire rebuild operation exception safe.
-    new_compacted_but_not_deleted.insert(new_compacted_but_not_deleted.end(), desc.input_sstables.begin(), desc.input_sstables.end());
+    new_compacted_but_not_deleted.insert(new_compacted_but_not_deleted.end(), desc.old_sstables.begin(), desc.old_sstables.end());
 
     _cache.invalidate([this, &desc] () noexcept {
         // FIXME: this is not really noexcept, but we need to provide strong exception guarantees.
-        rebuild_sstable_list(desc.output_sstables, desc.input_sstables);
+        rebuild_sstable_list(desc.new_sstables, desc.old_sstables);
     }, std::move(desc.ranges_for_cache_invalidation)).get();
 
     // refresh underlying data source in row cache to prevent it from holding reference
@@ -1222,7 +1222,7 @@ table::on_compaction_completion(sstables::compaction_completion_desc& desc) {
 
     rebuild_statistics();
 
-    auto f = seastar::with_gate(_sstable_deletion_gate, [this, sstables_to_remove = desc.input_sstables] {
+    auto f = seastar::with_gate(_sstable_deletion_gate, [this, sstables_to_remove = desc.old_sstables] {
        return with_semaphore(_sstable_deletion_sem, 1, [sstables_to_remove = std::move(sstables_to_remove)] {
            return sstables::delete_atomically(std::move(sstables_to_remove));
        });
@@ -1240,7 +1240,7 @@ table::on_compaction_completion(sstables::compaction_completion_desc& desc) {
     // or they could stay forever in the set, resulting in deleted files remaining
     // opened and disk space not being released until shutdown.
     std::unordered_set<sstables::shared_sstable> s(
-           desc.input_sstables.begin(), desc.input_sstables.end());
+           desc.old_sstables.begin(), desc.old_sstables.end());
     auto e = boost::range::remove_if(_sstables_compacted_but_not_deleted, [&] (sstables::shared_sstable sst) -> bool {
         return s.count(sst);
     });
@@ -1294,21 +1294,21 @@ table::compact_sstables(sstables::compaction_descriptor descriptor) {
     }
 
     return with_lock(_sstables_lock.for_read(), [this, descriptor = std::move(descriptor)] () mutable {
-        auto create_sstable = [this] {
+        descriptor.creator = [this] (shard_id dummy) {
                 auto sst = make_sstable();
                 sst->set_unshared();
                 return sst;
         };
-        auto replace_sstables = [this, release_exhausted = descriptor.release_exhausted] (sstables::compaction_completion_desc desc) {
-            _compaction_strategy.notify_completion(desc.input_sstables, desc.output_sstables);
-            _compaction_manager.propagate_replacement(this, desc.input_sstables, desc.output_sstables);
+        descriptor.replacer = [this, release_exhausted = descriptor.release_exhausted] (sstables::compaction_completion_desc desc) {
+            _compaction_strategy.notify_completion(desc.old_sstables, desc.new_sstables);
+            _compaction_manager.propagate_replacement(this, desc.old_sstables, desc.new_sstables);
             this->on_compaction_completion(desc);
             if (release_exhausted) {
-                release_exhausted(desc.input_sstables);
+                release_exhausted(desc.old_sstables);
             }
         };
 
-        return sstables::compact_sstables(std::move(descriptor), *this, create_sstable, replace_sstables);
+        return sstables::compact_sstables(std::move(descriptor), *this);
     }).then([this] (auto info) {
         if (info.type != sstables::compaction_type::Compaction) {
             return make_ready_future<>();
@@ -1437,7 +1437,7 @@ future<std::unordered_set<sstring>> table::get_sstables_by_partition_key(const s
             [this] (std::unordered_set<sstring>& filenames, lw_shared_ptr<sstables::sstable_set::incremental_selector>& sel, partition_key& pk) {
         return do_with(dht::decorated_key(dht::decorate_key(*_schema, pk)),
                 [this, &filenames, &sel, &pk](dht::decorated_key& dk) mutable {
-            auto sst = sel->select(dk).sstables;
+            const auto& sst = sel->select(dk).sstables;
             auto hk = sstables::sstable::make_hashed_key(*_schema, dk.key());
 
             return do_for_each(sst, [this, &filenames, &dk, hk = std::move(hk)] (std::vector<sstables::shared_sstable>::const_iterator::reference s) mutable {
@@ -1466,7 +1466,7 @@ std::vector<sstables::shared_sstable> table::select_sstables(const dht::partitio
     return _sstables->select(range);
 }
 
-std::vector<sstables::shared_sstable> table::candidates_for_compaction() const {
+std::vector<sstables::shared_sstable> table::non_staging_sstables() const {
     return boost::copy_range<std::vector<sstables::shared_sstable>>(*get_sstables()
             | boost::adaptors::filtered([this] (auto& sst) {
         return !_sstables_need_rewrite.count(sst->generation()) && !_sstables_staging.count(sst->generation());
@@ -2034,6 +2034,11 @@ void table::set_schema(schema_ptr s) {
     }
     _schema = std::move(s);
 
+    for (auto&& v : _views) {
+        v->view_info()->set_base_info(
+            v->view_info()->make_base_dependent_view_info(*_schema));
+    }
+
     set_compaction_strategy(_schema->compaction_strategy());
     trigger_compaction();
 }
@@ -2045,7 +2050,8 @@ static std::vector<view_ptr>::iterator find_view(std::vector<view_ptr>& views, c
 }
 
 void table::add_or_update_view(view_ptr v) {
-    v->view_info()->initialize_base_dependent_fields(*schema());
+    v->view_info()->set_base_info(
+        v->view_info()->make_base_dependent_view_info(*_schema));
     auto existing = find_view(_views, v);
     if (existing != _views.end()) {
         *existing = std::move(v);
@@ -2098,7 +2104,7 @@ static size_t memory_usage_of(const std::vector<frozen_mutation_and_schema>& ms)
  * @return a future resolving to the mutations to apply to the views, which can be empty.
  */
 future<> table::generate_and_propagate_view_updates(const schema_ptr& base,
-        std::vector<view_ptr>&& views,
+        std::vector<db::view::view_and_base>&& views,
         mutation&& m,
         flat_mutation_reader_opt existings) const {
     auto base_token = m.token();
@@ -2206,7 +2212,7 @@ table::local_base_lock(
  * @return a future that resolves when the updates have been acknowledged by the view replicas
  */
 future<> table::populate_views(
-        std::vector<view_ptr> views,
+        std::vector<db::view::view_and_base> views,
         dht::token base_token,
         flat_mutation_reader&& reader) {
     auto& schema = reader.schema();
@@ -2527,7 +2533,7 @@ future<row_locker::lock_holder> table::do_push_view_replica_updates(const schema
     }
     auto& base = schema();
     m.upgrade(base);
-    auto views = affected_views(base, m);
+    auto views = db::view::with_base_info_snapshot(affected_views(base, m));
     if (views.empty()) {
         return make_ready_future<row_locker::lock_holder>();
     }

test/alternator/run

@@ -28,8 +28,8 @@ fi
 SCYLLA_IP=127.1.$(($$ >> 8 & 255)).$(($$ & 255))
 echo "Running Scylla on $SCYLLA_IP"
 
-tmp_dir=/tmp/alternator-test-$$
-mkdir $tmp_dir
+tmp_dir="$(readlink -e ${TMPDIR-/tmp})"/alternator-test-$$
+mkdir "$tmp_dir"
 
 # We run the cleanup() function on exit for any reason - successful finish
 # of the script, an error (since we have "set -e"), or a signal.

test/alternator/test_condition_expression.py

@@ -1351,3 +1351,37 @@ def test_condition_expression_with_forbidden_rmw(scylla_only, dynamodb, test_tab
     assert test_table_s.get_item(Key={'p': s}, ConsistentRead=True)['Item'] == {'p': s, 'regular': 'write'}
     test_table_s.update_item(Key={'p': s}, AttributeUpdates={'write': {'Value': 'regular', 'Action': 'PUT'}})
     assert test_table_s.get_item(Key={'p': s}, ConsistentRead=True)['Item'] == {'p': s, 'regular': 'write', 'write': 'regular'}
+
+# Reproducer for issue #6573: binary strings should be ordered as unsigned
+# bytes, i.e., byte 128 comes after 127, not before as with signed bytes.
+# Test the five ordering operators: <, <=, >, >=, between
+def test_condition_expression_unsigned_bytes(test_table_s):
+    p = random_string()
+    test_table_s.put_item(Item={'p': p, 'b': bytearray([127])})
+    test_table_s.update_item(Key={'p': p},
+        UpdateExpression='SET z = :newval',
+        ConditionExpression='b < :oldval',
+        ExpressionAttributeValues={':newval': 1, ':oldval': bytearray([128])})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 1
+    test_table_s.update_item(Key={'p': p},
+        UpdateExpression='SET z = :newval',
+        ConditionExpression='b <= :oldval',
+        ExpressionAttributeValues={':newval': 2, ':oldval': bytearray([128])})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 2
+    test_table_s.update_item(Key={'p': p},
+        UpdateExpression='SET z = :newval',
+        ConditionExpression='b between :oldval1 and :oldval2',
+        ExpressionAttributeValues={':newval': 3, ':oldval1': bytearray([126]), ':oldval2': bytearray([128])})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 3
+
+    test_table_s.put_item(Item={'p': p, 'b': bytearray([128])})
+    test_table_s.update_item(Key={'p': p},
+        UpdateExpression='SET z = :newval',
+        ConditionExpression='b > :oldval',
+        ExpressionAttributeValues={':newval': 4, ':oldval': bytearray([127])})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 4
+    test_table_s.update_item(Key={'p': p},
+        UpdateExpression='SET z = :newval',
+        ConditionExpression='b >= :oldval',
+        ExpressionAttributeValues={':newval': 5, ':oldval': bytearray([127])})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 5

test/alternator/test_expected.py

@@ -1077,3 +1077,42 @@ def test_put_item_expected(test_table_s):
     assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item'] == {'p': p, 'a': 2}
     with pytest.raises(ClientError, match='ConditionalCheckFailedException'):
         test_table_s.put_item(Item={'p': p, 'a': 3}, Expected={'a': {'Value': 1}})
+
+# Reproducer for issue #6573: binary strings should be ordered as unsigned
+# bytes, i.e., byte 128 comes after 127, not before as with signed bytes.
+# Test the five ordering operators: LT, LE, GT, GE, BETWEEN
+def test_update_expected_unsigned_bytes(test_table_s):
+    p = random_string()
+    test_table_s.put_item(Item={'p': p, 'b': bytearray([127])})
+    test_table_s.update_item(Key={'p': p},
+        AttributeUpdates={'z': {'Value': 1, 'Action': 'PUT'}},
+        Expected={'b': {'ComparisonOperator': 'LT',
+                        'AttributeValueList': [bytearray([128])]}}
+    )
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 1
+    test_table_s.update_item(Key={'p': p},
+        AttributeUpdates={'z': {'Value': 2, 'Action': 'PUT'}},
+        Expected={'b': {'ComparisonOperator': 'LE',
+                        'AttributeValueList': [bytearray([128])]}}
+    )
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 2
+    test_table_s.update_item(Key={'p': p},
+        AttributeUpdates={'z': {'Value': 3, 'Action': 'PUT'}},
+        Expected={'b': {'ComparisonOperator': 'BETWEEN',
+                        'AttributeValueList': [bytearray([126]), bytearray([128])]}}
+    )
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 3
+
+    test_table_s.put_item(Item={'p': p, 'b': bytearray([128])})
+    test_table_s.update_item(Key={'p': p},
+        AttributeUpdates={'z': {'Value': 4, 'Action': 'PUT'}},
+        Expected={'b': {'ComparisonOperator': 'GT',
+                        'AttributeValueList': [bytearray([127])]}}
+    )
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 4
+    test_table_s.update_item(Key={'p': p},
+        AttributeUpdates={'z': {'Value': 5, 'Action': 'PUT'}},
+        Expected={'b': {'ComparisonOperator': 'GE',
+                        'AttributeValueList': [bytearray([127])]}}
+    )
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True)['Item']['z'] == 5

test/alternator/test_key_condition_expression.py

@@ -520,6 +520,42 @@ def test_key_condition_expression_and_conditions(test_table_sn_with_sorted_parti
                 'ComparisonOperator': 'GT'}}
             )
 
+# Demonstrate that issue #6573 was not a bug for KeyConditionExpression:
+# binary strings are ordered as unsigned bytes, i.e., byte 128 comes after
+# 127, not as signed bytes.
+# Test the five ordering operators: <, <=, >, >=, between
+def test_key_condition_expression_unsigned_bytes(test_table_sb):
+    p = random_string()
+    items = [{'p': p, 'c': bytearray([i])} for i in range(126,129)]
+    with test_table_sb.batch_writer() as batch:
+        for item in items:
+            batch.put_item(item)
+    got_items = full_query(test_table_sb,
+        KeyConditionExpression='p=:p AND c<:c',
+        ExpressionAttributeValues={':p': p, ':c': bytearray([127])})
+    expected_items = [item for item in items if item['c'] < bytearray([127])]
+    assert(got_items == expected_items)
+    got_items = full_query(test_table_sb,
+        KeyConditionExpression='p=:p AND c<=:c',
+        ExpressionAttributeValues={':p': p, ':c': bytearray([127])})
+    expected_items = [item for item in items if item['c'] <= bytearray([127])]
+    assert(got_items == expected_items)
+    got_items = full_query(test_table_sb,
+        KeyConditionExpression='p=:p AND c>:c',
+        ExpressionAttributeValues={':p': p, ':c': bytearray([127])})
+    expected_items = [item for item in items if item['c'] > bytearray([127])]
+    assert(got_items == expected_items)
+    got_items = full_query(test_table_sb,
+        KeyConditionExpression='p=:p AND c>=:c',
+        ExpressionAttributeValues={':p': p, ':c': bytearray([127])})
+    expected_items = [item for item in items if item['c'] >= bytearray([127])]
+    assert(got_items == expected_items)
+    got_items = full_query(test_table_sb,
+        KeyConditionExpression='p=:p AND c BETWEEN :c1 AND :c2',
+        ExpressionAttributeValues={':p': p, ':c1': bytearray([127]), ':c2': bytearray([128])})
+    expected_items = [item for item in items if item['c'] >= bytearray([127]) and item['c'] <= bytearray([128])]
+    assert(got_items == expected_items)
+
 # The following is an older test we had, which test one arbitrary use case
 # for KeyConditionExpression. It uses filled_test_table (the one we also
 # use in test_scan.py) instead of the fixtures defined in this file.

test/boost/big_decimal_test.cc

@@ -132,6 +132,13 @@ BOOST_AUTO_TEST_CASE(test_big_decimal_div) {
     test_div("-0.25", 10, "-0.02");
     test_div("-0.26", 10, "-0.03");
     test_div("-10E10", 3, "-3E10");
+
+    // Document a small oddity, 1e1 has -1 decimal places, so dividing
+    // it by 2 produces 0. This is not the behavior in cassandra, but
+    // scylla doesn't expose arithmetic operations, so this doesn't
+    // seem to be visible from CQL.
+    test_div("10", 2, "5");
+    test_div("1e1", 2, "0e1");
 }
 
 BOOST_AUTO_TEST_CASE(test_big_decimal_assignadd) {

test/boost/castas_fcts_test.cc

@@ -142,6 +142,19 @@ SEASTAR_TEST_CASE(test_decimal_to_bigint) {
     });
 }
 
+SEASTAR_TEST_CASE(test_decimal_to_float) {
+    return do_with_cql_env_thread([&](auto& e) {
+        e.execute_cql("CREATE TABLE test (key text primary key, value decimal)").get();
+        e.execute_cql("INSERT INTO test (key, value) VALUES ('k1', 10)").get();
+        e.execute_cql("INSERT INTO test (key, value) VALUES ('k2', 1e1)").get();
+        auto v = e.execute_cql("SELECT key, CAST(value as float) from test").get0();
+        assert_that(v).is_rows().with_rows_ignore_order({
+            {{serialized("k1")}, {serialized(float(10))}},
+            {{serialized("k2")}, {serialized(float(10))}},
+        });
+    });
+}
+
 SEASTAR_TEST_CASE(test_varint_to_bigint) {
     return do_with_cql_env_thread([&](auto& e) {
         e.execute_cql("CREATE TABLE test (key text primary key, value varint)").get();

test/boost/cql_query_test.cc

@@ -20,6 +20,7 @@
  */
 
 
+#include <boost/algorithm/string/join.hpp>
 #include <boost/range/irange.hpp>
 #include <boost/range/adaptors.hpp>
 #include <boost/range/algorithm.hpp>
@@ -1625,13 +1626,13 @@ SEASTAR_TEST_CASE(test_table_compression) {
         BOOST_REQUIRE(e.local_db().find_schema("ks", "tb5")->get_compressor_params().get_compressor() == nullptr);
 
         BOOST_REQUIRE_THROW(e.execute_cql(
-                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LossyCompressor' };"),
+                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LossyCompressor' };").get(),
                 std::exception);
         BOOST_REQUIRE_THROW(e.execute_cql(
-                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LZ4Compressor', 'chunk_length_kb' : -1 };"),
+                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LZ4Compressor', 'chunk_length_kb' : -1 };").get(),
                 std::exception);
         BOOST_REQUIRE_THROW(e.execute_cql(
-                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LZ4Compressor', 'chunk_length_kb' : 3 };"),
+                "create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LZ4Compressor', 'chunk_length_kb' : 3 };").get(),
                 std::exception);
 
         e.execute_cql("create table tb2 (foo text PRIMARY KEY, bar text) with compression = { 'sstable_compression' : 'LZ4Compressor', 'chunk_length_kb' : 2 };").get();
@@ -2717,6 +2718,85 @@ SEASTAR_TEST_CASE(test_reversed_slice_with_empty_range_before_all_rows) {
     });
 }
 
+// Test that the sstable layer correctly handles reversed slices, in particular
+// slices that read many clustering ranges, such that there is a large enough gap
+// between the ranges for the reader to attempt to use the promoted index for
+// skipping between them.
+// For this reason, the test writes a large partition (10MB), then issues a
+// reverse query which reads 4 singular clustering ranges from it. The ranges are
+// constructed such that there is many clustering rows between them: roughly 20%
+// which is ~2MB.
+// See #6171
+SEASTAR_TEST_CASE(test_reversed_slice_with_many_clustering_ranges) {
+    cql_test_config cfg;
+    cfg.db_config->max_memory_for_unlimited_query(std::numeric_limits<uint64_t>::max());
+    return do_with_cql_env_thread([] (cql_test_env& e) {
+        e.execute_cql("CREATE TABLE test (pk int, ck int, v text, PRIMARY KEY (pk, ck));").get();
+        auto id = e.prepare("INSERT INTO test (pk, ck, v) VALUES (?, ?, ?);").get0();
+
+        const int pk = 0;
+        const auto raw_pk = int32_type->decompose(data_value(pk));
+        const auto cql3_pk = cql3::raw_value::make_value(raw_pk);
+
+        const auto value = sstring(1024, 'a');
+        const auto raw_value = utf8_type->decompose(data_value(value));
+        const auto cql3_value = cql3::raw_value::make_value(raw_value);
+
+        const int num_rows = 10 * 1024;
+
+        for (int i = 0; i != num_rows; ++i) {
+            const auto cql3_ck = cql3::raw_value::make_value(int32_type->decompose(data_value(i)));
+            e.execute_prepared(id, {cql3_pk, cql3_ck, cql3_value}).get();
+        }
+
+        e.db().invoke_on_all([] (database& db) {
+            return db.flush_all_memtables();
+        }).get();
+
+        const std::vector<int> selected_cks{
+                2 * (num_rows / 10),
+                4 * (num_rows / 10),
+                6 * (num_rows / 10),
+                8 * (num_rows / 10)};
+
+        const auto make_expected_row = [&] (int ck) -> std::vector<bytes_opt> {
+            return {raw_pk, int32_type->decompose(ck), raw_value};
+        };
+
+        // Many singular ranges - to check that the right range is used for
+        // determining the disk read-range upper bound.
+        {
+            const auto select_query = format(
+                    "SELECT * FROM test WHERE pk = {} and ck IN ({}) ORDER BY ck DESC BYPASS CACHE;",
+                    pk,
+                    boost::algorithm::join(selected_cks | boost::adaptors::transformed([] (int ck) { return format("{}", ck); }), ", "));
+            assert_that(e.execute_cql(select_query).get0())
+                    .is_rows()
+                    .with_rows(boost::copy_range<std::vector<std::vector<bytes_opt>>>(
+                                selected_cks
+                                | boost::adaptors::reversed
+                                | boost::adaptors::transformed(make_expected_row)));
+        }
+
+        // A single wide range - to check that the right range bound is used for
+        // determining the disk read-range upper bound.
+        {
+            const auto select_query = format(
+                    "SELECT * FROM test WHERE pk = {} and ck >= {} and ck <= {} ORDER BY ck DESC BYPASS CACHE;",
+                    pk,
+                    selected_cks[0],
+                    selected_cks[1]);
+
+            assert_that(e.execute_cql(select_query).get0())
+                    .is_rows()
+                    .with_rows(boost::copy_range<std::vector<std::vector<bytes_opt>>>(
+                                boost::irange(selected_cks[0], selected_cks[1] + 1)
+                                | boost::adaptors::reversed
+                                | boost::adaptors::transformed(make_expected_row)));
+        }
+    }, std::move(cfg));
+}
+
 SEASTAR_TEST_CASE(test_query_with_range_tombstones) {
     return do_with_cql_env([] (cql_test_env& e) {
         return seastar::async([&e] {
@@ -3363,10 +3443,13 @@ SEASTAR_TEST_CASE(test_select_with_mixed_order_table) {
 }
 
 uint64_t
-run_and_examine_cache_stat_change(cql_test_env& e, uint64_t cache_tracker::stats::*metric, std::function<void (cql_test_env& e)> func) {
+run_and_examine_cache_read_stats_change(cql_test_env& e, std::string_view cf_name, std::function<void (cql_test_env& e)> func) {
     auto read_stat = [&] {
-        auto local_read_metric = [metric] (database& db) { return db.row_cache_tracker().get_stats().*metric; };
-        return e.db().map_reduce0(local_read_metric, uint64_t(0), std::plus<uint64_t>()).get0();
+        return e.db().map_reduce0([&cf_name] (const database& db) {
+            auto& t = db.find_column_family("ks", sstring(cf_name));
+            auto& stats = t.get_row_cache().stats();
+            return stats.reads_with_misses.count() + stats.reads_with_no_misses.count();
+        }, uint64_t(0), std::plus<uint64_t>()).get0();
     };
     auto before = read_stat();
     func(e);
@@ -3377,11 +3460,11 @@ run_and_examine_cache_stat_change(cql_test_env& e, uint64_t cache_tracker::stats
 SEASTAR_TEST_CASE(test_cache_bypass) {
     return do_with_cql_env_thread([] (cql_test_env& e) {
         e.execute_cql("CREATE TABLE t (k int PRIMARY KEY)").get();
-        auto with_cache = run_and_examine_cache_stat_change(e, &cache_tracker::stats::reads, [] (cql_test_env& e) {
+        auto with_cache = run_and_examine_cache_read_stats_change(e, "t", [] (cql_test_env& e) {
             e.execute_cql("SELECT * FROM t").get();
         });
         BOOST_REQUIRE(with_cache >= smp::count);  // scan may make multiple passes per shard
-        auto without_cache = run_and_examine_cache_stat_change(e, &cache_tracker::stats::reads, [] (cql_test_env& e) {
+        auto without_cache = run_and_examine_cache_read_stats_change(e, "t", [] (cql_test_env& e) {
             e.execute_cql("SELECT * FROM t BYPASS CACHE").get();
         });
         BOOST_REQUIRE_EQUAL(without_cache, 0);
@@ -4414,3 +4497,41 @@ SEASTAR_TEST_CASE(test_range_deletions_for_specific_column) {
                 exceptions::invalid_request_exception);
     });
 }
+
+SEASTAR_TEST_CASE(test_alter_table_default_ttl_reset) {
+    return do_with_cql_env([](cql_test_env& e) {
+        return seastar::async([&e] {
+            e.execute_cql("CREATE TABLE tbl (a int, b int, PRIMARY KEY (a)) WITH default_time_to_live=10").get();
+            BOOST_REQUIRE(e.local_db().find_schema("ks", "tbl")->default_time_to_live().count() == 10);
+            e.execute_cql("ALTER TABLE tbl WITH gc_grace_seconds=0").get();
+            BOOST_REQUIRE(e.local_db().find_schema("ks", "tbl")->default_time_to_live().count() == 10);
+        });
+    });
+}
+
+SEASTAR_TEST_CASE(test_impossible_where) {
+    return do_with_cql_env_thread([] (cql_test_env& e) {
+        cquery_nofail(e, "CREATE TABLE t(p int PRIMARY KEY, r int)");
+        cquery_nofail(e, "INSERT INTO  t(p,r) VALUES (0, 0)");
+        cquery_nofail(e, "INSERT INTO  t(p,r) VALUES (1, 10)");
+        cquery_nofail(e, "INSERT INTO  t(p,r) VALUES (2, 20)");
+        require_rows(e, "SELECT * FROM t WHERE r>10 AND r<10 ALLOW FILTERING", {});
+        require_rows(e, "SELECT * FROM t WHERE r>=10 AND r<=0 ALLOW FILTERING", {});
+
+        cquery_nofail(e, "CREATE TABLE t2(p int, c int, PRIMARY KEY(p, c)) WITH CLUSTERING ORDER BY (c DESC)");
+        cquery_nofail(e, "INSERT INTO  t2(p,c) VALUES (0, 0)");
+        cquery_nofail(e, "INSERT INTO  t2(p,c) VALUES (1, 10)");
+        cquery_nofail(e, "INSERT INTO  t2(p,c) VALUES (2, 20)");
+        require_rows(e, "SELECT * FROM t2 WHERE c>10 AND c<10 ALLOW FILTERING", {});
+        require_rows(e, "SELECT * FROM t2 WHERE c>=10 AND c<=0 ALLOW FILTERING", {});
+    });
+}
+
+SEASTAR_TEST_CASE(test_counter_column_added_into_non_counter_table) {
+    return do_with_cql_env_thread([] (cql_test_env& e) {
+        cquery_nofail(e, "CREATE TABLE t (pk int, ck int, PRIMARY KEY(pk, ck))");
+
+        BOOST_REQUIRE_THROW(e.execute_cql("ALTER TABLE t ADD \"c\" counter;").get(),
+                exceptions::configuration_exception);
+    });
+}

test/boost/filtering_test.cc

@@ -1134,6 +1134,9 @@ SEASTAR_TEST_CASE(test_filtering) {
                 { int32_type->decompose(8), int32_type->decompose(3) },
                 { int32_type->decompose(9), int32_type->decompose(3) },
             });
+            require_rows(e, "SELECT k FROM cf WHERE k=12 AND (m,n)>=(4,0) ALLOW FILTERING;", {
+                    { int32_type->decompose(12), int32_type->decompose(4), int32_type->decompose(5)},
+                });
         }
 
         // test filtering on clustering keys

test/boost/mutation_reader_test.cc

@@ -43,6 +43,7 @@
 #include "test/lib/make_random_string.hh"
 #include "test/lib/dummy_partitioner.hh"
 #include "test/lib/reader_lifecycle_policy.hh"
+#include "test/lib/random_utils.hh"
 
 #include "dht/sharder.hh"
 #include "mutation_reader.hh"
@@ -2737,3 +2738,597 @@ SEASTAR_THREAD_TEST_CASE(test_compacting_reader_next_partition) {
     }
     reader_assertions.produces_end_of_stream();
 }
+
+SEASTAR_THREAD_TEST_CASE(test_auto_paused_evictable_reader_is_mutation_source) {
+    auto make_populate = [] (schema_ptr s, const std::vector<mutation>& mutations, gc_clock::time_point query_time) {
+        auto mt = make_lw_shared<memtable>(s);
+        for (auto& mut : mutations) {
+            mt->apply(mut);
+        }
+        auto sem = make_lw_shared<reader_concurrency_semaphore>(reader_concurrency_semaphore::no_limits());
+        return mutation_source([=] (
+                schema_ptr s,
+                reader_permit permit,
+                const dht::partition_range& range,
+                const query::partition_slice& slice,
+                const io_priority_class& pc,
+                tracing::trace_state_ptr trace_state,
+                streamed_mutation::forwarding fwd_sm,
+                mutation_reader::forwarding fwd_mr) mutable {
+            auto mr = make_auto_paused_evictable_reader(mt->as_data_source(), std::move(s), *sem, range, slice, pc, std::move(trace_state), fwd_mr);
+            if (fwd_sm == streamed_mutation::forwarding::yes) {
+                return make_forwardable(std::move(mr));
+            }
+            return mr;
+        });
+    };
+
+    run_mutation_source_tests(make_populate);
+}
+
+SEASTAR_THREAD_TEST_CASE(test_manual_paused_evictable_reader_is_mutation_source) {
+    class maybe_pausing_reader : public flat_mutation_reader::impl {
+        flat_mutation_reader _reader;
+        std::optional<evictable_reader_handle> _handle;
+
+    private:
+        void maybe_pause() {
+            if (!tests::random::get_int(0, 4)) {
+                _handle->pause();
+            }
+        }
+
+    public:
+        maybe_pausing_reader(
+                memtable& mt,
+                reader_concurrency_semaphore& semaphore,
+                const dht::partition_range& pr,
+                const query::partition_slice& ps,
+                const io_priority_class& pc,
+                tracing::trace_state_ptr trace_state,
+                mutation_reader::forwarding fwd_mr)
+            : impl(mt.schema()), _reader(nullptr) {
+            std::tie(_reader, _handle) = make_manually_paused_evictable_reader(mt.as_data_source(), mt.schema(), semaphore, pr, ps, pc,
+                    std::move(trace_state), fwd_mr);
+        }
+        virtual future<> fill_buffer(db::timeout_clock::time_point timeout) override {
+            return _reader.fill_buffer(timeout).then([this] {
+                _end_of_stream = _reader.is_end_of_stream();
+                _reader.move_buffer_content_to(*this);
+            }).then([this] {
+                maybe_pause();
+            });
+        }
+        virtual void next_partition() override {
+            clear_buffer_to_next_partition();
+            if (!is_buffer_empty()) {
+                return;
+            }
+            _end_of_stream = false;
+            _reader.next_partition();
+        }
+        virtual future<> fast_forward_to(const dht::partition_range& pr, db::timeout_clock::time_point timeout) override {
+            clear_buffer();
+            _end_of_stream = false;
+            return _reader.fast_forward_to(pr, timeout).then([this] {
+                maybe_pause();
+            });
+        }
+        virtual future<> fast_forward_to(position_range pr, db::timeout_clock::time_point timeout) override {
+            throw_with_backtrace<std::bad_function_call>();
+        }
+        virtual size_t buffer_size() const override {
+            return flat_mutation_reader::impl::buffer_size() + _reader.buffer_size();
+        }
+    };
+
+    auto make_populate = [this] (schema_ptr s, const std::vector<mutation>& mutations, gc_clock::time_point query_time) {
+        auto mt = make_lw_shared<memtable>(s);
+        for (auto& mut : mutations) {
+            mt->apply(mut);
+        }
+        auto sem = make_lw_shared<reader_concurrency_semaphore>(reader_concurrency_semaphore::no_limits());
+        return mutation_source([=] (
+                schema_ptr s,
+                reader_permit permit,
+                const dht::partition_range& range,
+                const query::partition_slice& slice,
+                const io_priority_class& pc,
+                tracing::trace_state_ptr trace_state,
+                streamed_mutation::forwarding fwd_sm,
+                mutation_reader::forwarding fwd_mr) mutable {
+            auto mr = make_flat_mutation_reader<maybe_pausing_reader>(*mt, *sem, range, slice, pc, std::move(trace_state), fwd_mr);
+            if (fwd_sm == streamed_mutation::forwarding::yes) {
+                return make_forwardable(std::move(mr));
+            }
+            return mr;
+        });
+    };
+
+    run_mutation_source_tests(make_populate);
+}
+
+namespace {
+
+std::deque<mutation_fragment> copy_fragments(const schema& s, const std::deque<mutation_fragment>& o) {
+    std::deque<mutation_fragment> buf;
+    for (const auto& mf : o) {
+        buf.emplace_back(s, mf);
+    }
+    return buf;
+}
+
+flat_mutation_reader create_evictable_reader_and_evict_after_first_buffer(
+        schema_ptr schema,
+        reader_concurrency_semaphore& semaphore,
+        const dht::partition_range& prange,
+        const query::partition_slice& slice,
+        std::deque<mutation_fragment> first_buffer,
+        position_in_partition_view last_fragment_position,
+        std::deque<mutation_fragment> second_buffer,
+        size_t max_buffer_size) {
+    class factory {
+        schema_ptr _schema;
+        std::optional<std::deque<mutation_fragment>> _first_buffer;
+        std::optional<std::deque<mutation_fragment>> _second_buffer;
+        size_t _max_buffer_size;
+
+    private:
+        std::optional<std::deque<mutation_fragment>> copy_buffer(const std::optional<std::deque<mutation_fragment>>& o) {
+            if (!o) {
+                return {};
+            }
+            return copy_fragments(*_schema, *o);
+        }
+
+    public:
+        factory(schema_ptr schema, std::deque<mutation_fragment> first_buffer, std::deque<mutation_fragment> second_buffer, size_t max_buffer_size)
+            : _schema(std::move(schema)), _first_buffer(std::move(first_buffer)), _second_buffer(std::move(second_buffer)), _max_buffer_size(max_buffer_size) {
+        }
+
+        factory(const factory& o)
+            : _schema(o._schema)
+            , _first_buffer(copy_buffer(o._first_buffer))
+            , _second_buffer(copy_buffer(o._second_buffer)) {
+        }
+        factory(factory&& o) = default;
+
+        flat_mutation_reader operator()(
+                schema_ptr s,
+                reader_permit permit,
+                const dht::partition_range& range,
+                const query::partition_slice& slice,
+                const io_priority_class& pc,
+                tracing::trace_state_ptr trace_state,
+                streamed_mutation::forwarding fwd_sm,
+                mutation_reader::forwarding fwd_mr) {
+            BOOST_REQUIRE(s == _schema);
+            if (_first_buffer) {
+                auto buf = *std::exchange(_first_buffer, {});
+                auto rd = make_flat_mutation_reader_from_fragments(_schema, std::move(buf));
+                rd.set_max_buffer_size(_max_buffer_size);
+                return rd;
+            }
+            if (_second_buffer) {
+                auto buf = *std::exchange(_second_buffer, {});
+                auto rd = make_flat_mutation_reader_from_fragments(_schema, std::move(buf));
+                rd.set_max_buffer_size(_max_buffer_size);
+                return rd;
+            }
+            return make_empty_flat_reader(_schema);
+        }
+    };
+    auto ms = mutation_source(factory(schema, std::move(first_buffer), std::move(second_buffer), max_buffer_size));
+
+    auto [rd, handle] = make_manually_paused_evictable_reader(
+            std::move(ms),
+            schema,
+            semaphore,
+            prange,
+            slice,
+            seastar::default_priority_class(),
+            nullptr,
+            mutation_reader::forwarding::yes);
+
+    rd.set_max_buffer_size(max_buffer_size);
+
+    rd.fill_buffer(db::no_timeout).get0();
+
+    const auto eq_cmp = position_in_partition::equal_compare(*schema);
+    BOOST_REQUIRE(rd.is_buffer_full());
+    BOOST_REQUIRE(eq_cmp(rd.buffer().back().position(), last_fragment_position));
+    BOOST_REQUIRE(!rd.is_end_of_stream());
+
+    rd.detach_buffer();
+
+    handle.pause();
+
+    while(semaphore.try_evict_one_inactive_read());
+
+    return std::move(rd);
+}
+
+}
+
+SEASTAR_THREAD_TEST_CASE(test_evictable_reader_trim_range_tombstones) {
+    reader_concurrency_semaphore semaphore(reader_concurrency_semaphore::no_limits{});
+    simple_schema s;
+
+    const auto pkey = s.make_pkey();
+    size_t max_buffer_size = 512;
+    const int first_ck = 100;
+    const int second_buffer_ck = first_ck + 100;
+
+    size_t mem_usage = 0;
+
+    std::deque<mutation_fragment> first_buffer;
+    first_buffer.emplace_back(partition_start{pkey, {}});
+    mem_usage = first_buffer.back().memory_usage(*s.schema());
+    for (int i = 0; i < second_buffer_ck; ++i) {
+        first_buffer.emplace_back(s.make_row(s.make_ckey(i++), "v"));
+        mem_usage += first_buffer.back().memory_usage(*s.schema());
+    }
+    const auto last_fragment_position = position_in_partition(first_buffer.back().position());
+    max_buffer_size = mem_usage;
+    first_buffer.emplace_back(s.make_row(s.make_ckey(second_buffer_ck), "v"));
+
+    std::deque<mutation_fragment> second_buffer;
+    second_buffer.emplace_back(partition_start{pkey, {}});
+    mem_usage = second_buffer.back().memory_usage(*s.schema());
+    second_buffer.emplace_back(s.make_range_tombstone(query::clustering_range::make_ending_with(s.make_ckey(second_buffer_ck + 10))));
+    int ckey = second_buffer_ck;
+    while (mem_usage <= max_buffer_size) {
+        second_buffer.emplace_back(s.make_row(s.make_ckey(ckey++), "v"));
+        mem_usage += second_buffer.back().memory_usage(*s.schema());
+    }
+    second_buffer.emplace_back(partition_end{});
+
+    auto rd = create_evictable_reader_and_evict_after_first_buffer(s.schema(), semaphore, query::full_partition_range,
+            s.schema()->full_slice(), std::move(first_buffer), last_fragment_position, std::move(second_buffer), max_buffer_size);
+
+    rd.fill_buffer(db::no_timeout).get();
+
+    const auto tri_cmp = position_in_partition::tri_compare(*s.schema());
+
+    BOOST_REQUIRE(tri_cmp(last_fragment_position, rd.peek_buffer().position()) < 0);
+}
+
+namespace {
+
+void check_evictable_reader_validation_is_triggered(
+        std::string_view test_name,
+        std::string_view error_prefix, // empty str if no exception is expected
+        schema_ptr schema,
+        reader_concurrency_semaphore& semaphore,
+        const dht::partition_range& prange,
+        const query::partition_slice& slice,
+        std::deque<mutation_fragment> first_buffer,
+        position_in_partition_view last_fragment_position,
+        std::deque<mutation_fragment> second_buffer,
+        size_t max_buffer_size) {
+
+    testlog.info("check_evictable_reader_validation_is_triggered(): checking {} test case: {}", error_prefix.empty() ? "positive" : "negative", test_name);
+
+    auto rd = create_evictable_reader_and_evict_after_first_buffer(std::move(schema), semaphore, prange, slice, std::move(first_buffer),
+            last_fragment_position, std::move(second_buffer), max_buffer_size);
+
+    const bool fail = !error_prefix.empty();
+
+    try {
+        rd.fill_buffer(db::no_timeout).get0();
+    } catch (std::runtime_error& e) {
+        if (fail) {
+            if (error_prefix == std::string_view(e.what(), error_prefix.size())) {
+                testlog.trace("Expected exception caught: {}", std::current_exception());
+                return;
+            } else {
+                BOOST_FAIL(fmt::format("Exception with unexpected message caught: {}", std::current_exception()));
+            }
+        } else {
+            BOOST_FAIL(fmt::format("Unexpected exception caught: {}", std::current_exception()));
+        }
+    }
+    if (fail) {
+        BOOST_FAIL(fmt::format("Expected exception not thrown"));
+    }
+}
+
+}
+
+SEASTAR_THREAD_TEST_CASE(test_evictable_reader_self_validation) {
+    set_abort_on_internal_error(false);
+    auto reset_on_internal_abort = defer([] {
+        set_abort_on_internal_error(true);
+    });
+
+    reader_concurrency_semaphore semaphore(reader_concurrency_semaphore::no_limits{});
+    simple_schema s;
+
+    auto pkeys = s.make_pkeys(4);
+    boost::sort(pkeys, dht::decorated_key::less_comparator(s.schema()));
+
+    size_t max_buffer_size = 512;
+    const int first_ck = 100;
+    const int second_buffer_ck = first_ck + 100;
+    const int last_ck = second_buffer_ck + 100;
+
+    static const char partition_error_prefix[] = "maybe_validate_partition_start(): validation failed";
+    static const char position_in_partition_error_prefix[] = "validate_position_in_partition(): validation failed";
+    static const char trim_range_tombstones_error_prefix[] = "maybe_trim_range_tombstone(): validation failed";
+
+    const auto prange = dht::partition_range::make(
+            dht::partition_range::bound(pkeys[1], true),
+            dht::partition_range::bound(pkeys[2], true));
+
+    const auto ckrange = query::clustering_range::make(
+            query::clustering_range::bound(s.make_ckey(first_ck), true),
+            query::clustering_range::bound(s.make_ckey(last_ck), true));
+
+    const auto slice = partition_slice_builder(*s.schema()).with_range(ckrange).build();
+
+    std::deque<mutation_fragment> first_buffer;
+    first_buffer.emplace_back(partition_start{pkeys[1], {}});
+    size_t mem_usage = first_buffer.back().memory_usage(*s.schema());
+    for (int i = 0; i < second_buffer_ck; ++i) {
+        first_buffer.emplace_back(s.make_row(s.make_ckey(i++), "v"));
+        mem_usage += first_buffer.back().memory_usage(*s.schema());
+    }
+    max_buffer_size = mem_usage;
+    auto last_fragment_position = position_in_partition(first_buffer.back().position());
+    first_buffer.emplace_back(s.make_row(s.make_ckey(second_buffer_ck), "v"));
+
+    auto make_second_buffer = [&s, &max_buffer_size, second_buffer_ck] (dht::decorated_key pkey, std::optional<int> first_ckey = {},
+            bool inject_range_tombstone = false) mutable {
+        auto ckey = first_ckey ? *first_ckey : second_buffer_ck;
+        std::deque<mutation_fragment> second_buffer;
+        second_buffer.emplace_back(partition_start{std::move(pkey), {}});
+        size_t mem_usage = second_buffer.back().memory_usage(*s.schema());
+        if (inject_range_tombstone) {
+            second_buffer.emplace_back(s.make_range_tombstone(query::clustering_range::make_ending_with(s.make_ckey(last_ck))));
+        }
+        while (mem_usage <= max_buffer_size) {
+            second_buffer.emplace_back(s.make_row(s.make_ckey(ckey++), "v"));
+            mem_usage += second_buffer.back().memory_usage(*s.schema());
+        }
+        second_buffer.emplace_back(partition_end{});
+        return second_buffer;
+    };
+
+    //
+    // Continuing the same partition
+    //
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey < _last_pkey; pkey ∉ prange",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[0]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey",
+            "",
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∉ ckrange (<)",
+            position_in_partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], first_ck - 10),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∉ ckrange (<); start with trimmable range-tombstone",
+            position_in_partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], first_ck - 10, true),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∉ ckrange; position_in_partition < _next_position_in_partition",
+            position_in_partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], second_buffer_ck - 2),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∉ ckrange; position_in_partition < _next_position_in_partition; start with trimmable range-tombstone",
+            position_in_partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], second_buffer_ck - 2, true),
+            max_buffer_size);
+
+    {
+        auto second_buffer = make_second_buffer(pkeys[1], second_buffer_ck);
+        second_buffer[1] = s.make_range_tombstone(query::clustering_range::make_ending_with(s.make_ckey(second_buffer_ck - 10)));
+        check_evictable_reader_validation_is_triggered(
+                "pkey == _last_pkey; end(range_tombstone) < _next_position_in_partition",
+                trim_range_tombstones_error_prefix,
+                s.schema(),
+                semaphore,
+                prange,
+                slice,
+                copy_fragments(*s.schema(), first_buffer),
+                last_fragment_position,
+                std::move(second_buffer),
+                max_buffer_size);
+    }
+
+    {
+        auto second_buffer = make_second_buffer(pkeys[1], second_buffer_ck);
+        second_buffer[1] = s.make_range_tombstone(query::clustering_range::make_ending_with(s.make_ckey(second_buffer_ck + 10)));
+        check_evictable_reader_validation_is_triggered(
+                "pkey == _last_pkey; end(range_tombstone) > _next_position_in_partition",
+                "",
+                s.schema(),
+                semaphore,
+                prange,
+                slice,
+                copy_fragments(*s.schema(), first_buffer),
+                last_fragment_position,
+                std::move(second_buffer),
+                max_buffer_size);
+    }
+
+    {
+        auto second_buffer = make_second_buffer(pkeys[1], second_buffer_ck);
+        second_buffer[1] = s.make_range_tombstone(query::clustering_range::make_starting_with(s.make_ckey(last_ck + 10)));
+        check_evictable_reader_validation_is_triggered(
+                "pkey == _last_pkey; start(range_tombstone) ∉ ckrange (>)",
+                position_in_partition_error_prefix,
+                s.schema(),
+                semaphore,
+                prange,
+                slice,
+                copy_fragments(*s.schema(), first_buffer),
+                last_fragment_position,
+                std::move(second_buffer),
+                max_buffer_size);
+    }
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∈ ckrange",
+            "",
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], second_buffer_ck),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey; position_in_partition ∉ ckrange (>)",
+            position_in_partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1], last_ck + 10),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey > _last_pkey; pkey ∈ pkrange",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[2]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey > _last_pkey; pkey ∉ pkrange",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[3]),
+            max_buffer_size);
+
+    //
+    // Continuing from next partition
+    //
+
+    first_buffer.clear();
+
+    first_buffer.emplace_back(partition_start{pkeys[1], {}});
+    mem_usage = first_buffer.back().memory_usage(*s.schema());
+    for (int i = 0; i < second_buffer_ck; ++i) {
+        first_buffer.emplace_back(s.make_row(s.make_ckey(i++), "v"));
+        mem_usage += first_buffer.back().memory_usage(*s.schema());
+    }
+    first_buffer.emplace_back(partition_end{});
+    mem_usage += first_buffer.back().memory_usage(*s.schema());
+    last_fragment_position = position_in_partition(first_buffer.back().position());
+    max_buffer_size = mem_usage;
+    first_buffer.emplace_back(partition_start{pkeys[2], {}});
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey < _last_pkey; pkey ∉ pkrange",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[0]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey == _last_pkey",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[1]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey > _last_pkey; pkey ∈ pkrange",
+            "",
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[2]),
+            max_buffer_size);
+
+    check_evictable_reader_validation_is_triggered(
+            "pkey > _last_pkey; pkey ∉ pkrange",
+            partition_error_prefix,
+            s.schema(),
+            semaphore,
+            prange,
+            slice,
+            copy_fragments(*s.schema(), first_buffer),
+            last_fragment_position,
+            make_second_buffer(pkeys[3]),
+            max_buffer_size);
+}

test/boost/sstable_3_x_test.cc

@@ -2999,12 +2999,14 @@ static flat_mutation_reader compacted_sstable_reader(test_env& env, schema_ptr s
     tmpdir tmp;
     auto sstables = open_sstables(env, s, format("test/resource/sstables/3.x/uncompressed/{}", table_name), generations);
     auto new_generation = generations.back() + 1;
-    auto new_sstable = [s, &tmp, &env, new_generation] {
+
+    auto desc = sstables::compaction_descriptor(std::move(sstables));
+    desc.creator = [s, &tmp, &env, new_generation] (shard_id dummy) {
         return env.make_sstable(s, tmp.path().string(), new_generation,
                          sstables::sstable_version_types::mc, sstable::format_types::big, 4096);
     };
-
-    sstables::compact_sstables(sstables::compaction_descriptor(std::move(sstables)), *cf, new_sstable, replacer_fn_no_op()).get();
+    desc.replacer = replacer_fn_no_op();
+    sstables::compact_sstables(std::move(desc), *cf).get();
 
     auto compacted_sst = open_sstable(env, s, tmp.path().string(), new_generation);
     return compacted_sst->as_mutation_source().make_reader(s, no_reader_permit(), query::full_partition_range, s->full_slice());

test/boost/sstable_datafile_test.cc

@@ -1145,7 +1145,7 @@ SEASTAR_TEST_CASE(compact) {
                 return env.make_sstable(s, tmpdir_path,
                         (*gen)++, sstables::sstable::version_types::la, sstables::sstable::format_types::big);
             };
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(sstables)), *cf, new_sstable, replacer_fn_no_op()).then([&env, s, generation, cf, cm, tmpdir_path] (auto) {
+            return compact_sstables(sstables::compaction_descriptor(std::move(sstables)), *cf, new_sstable).then([&env, s, generation, cf, cm, tmpdir_path] (auto) {
                 // Verify that the compacted sstable has the right content. We expect to see:
                 //  name  | age | height
                 // -------+-----+--------
@@ -1291,7 +1291,7 @@ static future<std::vector<unsigned long>> compact_sstables(test_env& env, sstrin
             auto sstables_to_compact = sstables::size_tiered_compaction_strategy::most_interesting_bucket(*sstables, min_threshold, max_threshold);
             // We do expect that all candidates were selected for compaction (in this case).
             BOOST_REQUIRE(sstables_to_compact.size() == sstables->size());
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(sstables_to_compact)), *cf, new_sstable, replacer_fn_no_op()).then([generation] (auto) {});
+            return compact_sstables(sstables::compaction_descriptor(std::move(sstables_to_compact)), *cf, new_sstable).then([generation] (auto) {});
         } else if (strategy == compaction_strategy_type::leveled) {
             for (auto& sst : *sstables) {
                 BOOST_REQUIRE(sst->get_sstable_level() == 0);
@@ -1308,8 +1308,8 @@ static future<std::vector<unsigned long>> compact_sstables(test_env& env, sstrin
             BOOST_REQUIRE(candidate.level == 1);
             BOOST_REQUIRE(candidate.max_sstable_bytes == 1024*1024);
 
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(candidate.sstables), candidate.level, 1024*1024),
-                *cf, new_sstable, replacer_fn_no_op()).then([generation] (auto) {});
+            return compact_sstables(sstables::compaction_descriptor(std::move(candidate.sstables), candidate.level, 1024*1024),
+                *cf, new_sstable).then([generation] (auto) {});
         } else {
             throw std::runtime_error("unexpected strategy");
         }
@@ -2234,7 +2234,7 @@ SEASTAR_TEST_CASE(tombstone_purge_test) {
             for (auto&& sst : all) {
                 column_family_test(cf).add_sstable(sst);
             }
-            return sstables::compact_sstables(sstables::compaction_descriptor(to_compact), *cf, sst_gen, replacer_fn_no_op()).get0().new_sstables;
+            return compact_sstables(sstables::compaction_descriptor(to_compact), *cf, sst_gen).get0().new_sstables;
         };
 
         auto next_timestamp = [] {
@@ -2496,7 +2496,7 @@ SEASTAR_TEST_CASE(sstable_rewrite) {
             std::vector<shared_sstable> sstables;
             sstables.push_back(std::move(sstp));
 
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(sstables)), *cf, creator, replacer_fn_no_op()).then([s, key, new_tables] (auto) {
+            return compact_sstables(sstables::compaction_descriptor(std::move(sstables)), *cf, creator).then([s, key, new_tables] (auto) {
                 BOOST_REQUIRE(new_tables->size() == 1);
                 auto newsst = (*new_tables)[0];
                 BOOST_REQUIRE(newsst->generation() == 52);
@@ -2884,8 +2884,7 @@ SEASTAR_TEST_CASE(test_sstable_max_local_deletion_time_2) {
                 BOOST_REQUIRE(now.time_since_epoch().count() == sst2->get_stats_metadata().max_local_deletion_time);
 
                 auto creator = [&env, s, tmpdir_path, version, gen = make_lw_shared<unsigned>(56)] { return env.make_sstable(s, tmpdir_path, (*gen)++, version, big); };
-                auto info = sstables::compact_sstables(sstables::compaction_descriptor({sst1, sst2}), *cf,
-                                                       creator, replacer_fn_no_op()).get0();
+                auto info = compact_sstables(sstables::compaction_descriptor({sst1, sst2}), *cf, creator).get0();
                 BOOST_REQUIRE(info.new_sstables.size() == 1);
                 BOOST_REQUIRE(((now + gc_clock::duration(100)).time_since_epoch().count()) ==
                               info.new_sstables.front()->get_stats_metadata().max_local_deletion_time);
@@ -2982,7 +2981,7 @@ SEASTAR_TEST_CASE(compaction_with_fully_expired_table) {
         expired = get_fully_expired_sstables(*cf, ssts, gc_clock::now());
         BOOST_REQUIRE(expired.size() == 0);
 
-        auto ret = sstables::compact_sstables(sstables::compaction_descriptor(ssts), *cf, sst_gen, replacer_fn_no_op()).get0();
+        auto ret = compact_sstables(sstables::compaction_descriptor(ssts), *cf, sst_gen).get0();
         BOOST_REQUIRE(ret.start_size == sst->bytes_on_disk());
         BOOST_REQUIRE(ret.total_keys_written == 0);
         BOOST_REQUIRE(ret.new_sstables.empty());
@@ -3442,7 +3441,7 @@ SEASTAR_TEST_CASE(min_max_clustering_key_test_2) {
             check_min_max_column_names(sst2, {"9ck101"}, {"9ck298"});
 
             auto creator = [&env, s, &tmp, version] { return env.make_sstable(s, tmp.path().string(), 3, version, big); };
-            auto info = sstables::compact_sstables(sstables::compaction_descriptor({sst, sst2}), *cf, creator, replacer_fn_no_op()).get0();
+            auto info = compact_sstables(sstables::compaction_descriptor({sst, sst2}), *cf, creator).get0();
             BOOST_REQUIRE(info.new_sstables.size() == 1);
             check_min_max_column_names(info.new_sstables.front(), {"0ck100"}, {"9ck298"});
         }
@@ -4080,7 +4079,7 @@ SEASTAR_TEST_CASE(sstable_expired_data_ratio) {
             sst->set_unshared();
             return sst;
         };
-        auto info = sstables::compact_sstables(sstables::compaction_descriptor({ sst }), *cf, creator, replacer_fn_no_op()).get0();
+        auto info = compact_sstables(sstables::compaction_descriptor({ sst }), *cf, creator).get0();
         BOOST_REQUIRE(info.new_sstables.size() == 1);
         BOOST_REQUIRE(info.new_sstables.front()->estimate_droppable_tombstone_ratio(gc_before) == 0.0f);
         BOOST_REQUIRE_CLOSE(info.new_sstables.front()->data_size(), uncompacted_size*(1-expired), 5);
@@ -4351,8 +4350,8 @@ SEASTAR_TEST_CASE(compaction_correctness_with_partitioned_sstable_set) {
             // NEEDED for partitioned_sstable_set to actually have an effect
             std::for_each(all.begin(), all.end(), [] (auto& sst) { sst->set_sstable_level(1); });
             column_family_for_tests cf(s);
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(all), 0, 0 /*std::numeric_limits<uint64_t>::max()*/),
-                *cf, sst_gen, replacer_fn_no_op()).get0().new_sstables;
+            return compact_sstables(sstables::compaction_descriptor(std::move(all), 0, 0 /*std::numeric_limits<uint64_t>::max()*/),
+                *cf, sst_gen).get0().new_sstables;
         };
 
         auto make_insert = [&] (auto p) {
@@ -4591,7 +4590,7 @@ SEASTAR_TEST_CASE(sstable_cleanup_correctness_test) {
 
             auto descriptor = sstables::compaction_descriptor({std::move(sst)}, compaction_descriptor::default_level,
                 compaction_descriptor::default_max_sstable_bytes, run_identifier, compaction_options::make_cleanup());
-            auto ret = sstables::compact_sstables(std::move(descriptor), *cf, sst_gen, sstables::replacer_fn_no_op()).get0();
+            auto ret = compact_sstables(std::move(descriptor), *cf, sst_gen).get0();
 
             BOOST_REQUIRE(ret.total_keys_written == total_partitions);
             BOOST_REQUIRE(ret.new_sstables.size() == 1);
@@ -4734,8 +4733,8 @@ SEASTAR_TEST_CASE(sstable_scrub_test) {
 
             table->add_sstable_and_update_cache(sst).get();
 
-            BOOST_REQUIRE(table->candidates_for_compaction().size() == 1);
-            BOOST_REQUIRE(table->candidates_for_compaction().front() == sst);
+            BOOST_REQUIRE(table->non_staging_sstables().size() == 1);
+            BOOST_REQUIRE(table->non_staging_sstables().front() == sst);
 
             auto verify_fragments = [&] (sstables::shared_sstable sst, const std::vector<mutation_fragment>& mfs) {
                 auto r = assert_that(sst->as_mutation_source().make_reader(schema));
@@ -4756,7 +4755,7 @@ SEASTAR_TEST_CASE(sstable_scrub_test) {
             // We expect the scrub with skip_corrupted=false to stop on the first invalid fragment.
             compaction_manager.perform_sstable_scrub(table.get(), false).get();
 
-            BOOST_REQUIRE(table->candidates_for_compaction().size() == 1);
+            BOOST_REQUIRE(table->non_staging_sstables().size() == 1);
             verify_fragments(sst, corrupt_fragments);
 
             testlog.info("Scrub with --skip-corrupted=true");
@@ -4764,9 +4763,9 @@ SEASTAR_TEST_CASE(sstable_scrub_test) {
             // We expect the scrub with skip_corrupted=true to get rid of all invalid data.
             compaction_manager.perform_sstable_scrub(table.get(), true).get();
 
-            BOOST_REQUIRE(table->candidates_for_compaction().size() == 1);
-            BOOST_REQUIRE(table->candidates_for_compaction().front() != sst);
-            verify_fragments(table->candidates_for_compaction().front(), scrubbed_fragments);
+            BOOST_REQUIRE(table->non_staging_sstables().size() == 1);
+            BOOST_REQUIRE(table->non_staging_sstables().front() != sst);
+            verify_fragments(table->non_staging_sstables().front(), scrubbed_fragments);
         });
     }, test_cfg);
 }
@@ -4994,7 +4993,7 @@ SEASTAR_TEST_CASE(sstable_run_based_compaction_test) {
         cf->start();
         cf->set_compaction_strategy(sstables::compaction_strategy_type::size_tiered);
         auto compact = [&, s] (std::vector<shared_sstable> all, auto replacer) -> std::vector<shared_sstable> {
-            return sstables::compact_sstables(sstables::compaction_descriptor(std::move(all), 1, 0), *cf, sst_gen, replacer).get0().new_sstables;
+            return compact_sstables(sstables::compaction_descriptor(std::move(all), 1, 0), *cf, sst_gen, replacer).get0().new_sstables;
         };
         auto make_insert = [&] (auto p) {
             auto key = partition_key::from_exploded(*s, {to_bytes(p.first)});
@@ -5061,8 +5060,8 @@ SEASTAR_TEST_CASE(sstable_run_based_compaction_test) {
             auto expected_sst = sstable_run.begin();
             auto closed_sstables_tracker = sstable_run.begin();
             auto replacer = [&] (sstables::compaction_completion_desc desc) {
-                auto old_sstables = std::move(desc.input_sstables);
-                auto new_sstables = std::move(desc.output_sstables);
+                auto old_sstables = std::move(desc.old_sstables);
+                auto new_sstables = std::move(desc.new_sstables);
                 BOOST_REQUIRE(expected_sst != sstable_run.end());
                 if (incremental_enabled) {
                     do_incremental_replace(std::move(old_sstables), std::move(new_sstables), expected_sst, closed_sstables_tracker);
@@ -5218,7 +5217,7 @@ SEASTAR_TEST_CASE(backlog_tracker_correctness_after_stop_tracking_compaction) {
             // Start compaction, then stop tracking compaction, switch to TWCS, wait for compaction to finish and check for backlog.
             // That's done to assert backlog will work for compaction that is finished and was stopped tracking.
 
-            auto fut = sstables::compact_sstables(sstables::compaction_descriptor(ssts), *cf, sst_gen, replacer_fn_no_op());
+            auto fut = compact_sstables(sstables::compaction_descriptor(ssts), *cf, sst_gen);
 
             bool stopped_tracking = false;
             for (auto& info : cf._data->cm.get_compactions()) {
@@ -5537,9 +5536,10 @@ SEASTAR_TEST_CASE(incremental_compaction_data_resurrection_test) {
             return m;
         };
 
+        auto deletion_time = gc_clock::now();
         auto make_delete = [&] (partition_key key) {
             mutation m(s, key);
-            tombstone tomb(next_timestamp(), gc_clock::now());
+            tombstone tomb(next_timestamp(), deletion_time);
             m.partition().apply(tomb);
             return m;
         };
@@ -5597,15 +5597,20 @@ SEASTAR_TEST_CASE(incremental_compaction_data_resurrection_test) {
         BOOST_REQUIRE(is_partition_dead(alpha));
 
         auto replacer = [&] (sstables::compaction_completion_desc desc) {
-            auto old_sstables = std::move(desc.input_sstables);
-            auto new_sstables = std::move(desc.output_sstables);
+            auto old_sstables = std::move(desc.old_sstables);
+            auto new_sstables = std::move(desc.new_sstables);
             // expired_sst is exhausted, and new sstable is written with mut 2.
             BOOST_REQUIRE(old_sstables.size() == 1);
             BOOST_REQUIRE(old_sstables.front() == expired_sst);
-            BOOST_REQUIRE(new_sstables.size() == 1);
-            assert_that(sstable_reader(new_sstables.front(), s))
+            BOOST_REQUIRE(new_sstables.size() == 2);
+            for (auto& new_sstable : new_sstables) {
+                if (new_sstable->get_max_local_deletion_time() == deletion_time) { // Skipping GC SSTable.
+                    continue;
+                }
+                assert_that(sstable_reader(new_sstable, s))
                     .produces(mut2)
                     .produces_end_of_stream();
+            }
             column_family_test(cf).rebuild_sstable_list(new_sstables, old_sstables);
             // force compaction failure after sstable containing expired tombstone is removed from set.
             throw std::runtime_error("forcing compaction failure on early replacement");
@@ -5615,7 +5620,7 @@ SEASTAR_TEST_CASE(incremental_compaction_data_resurrection_test) {
         try {
             // The goal is to have one sstable generated for each mutation to trigger the issue.
             auto max_sstable_size = 0;
-            auto result = sstables::compact_sstables(sstables::compaction_descriptor(sstables, 0, max_sstable_size), *cf, sst_gen, replacer).get0().new_sstables;
+            auto result = compact_sstables(sstables::compaction_descriptor(sstables, 0, max_sstable_size), *cf, sst_gen, replacer).get0().new_sstables;
             BOOST_REQUIRE_EQUAL(2, result.size());
         } catch (...) {
             // swallow exception
@@ -5626,3 +5631,48 @@ SEASTAR_TEST_CASE(incremental_compaction_data_resurrection_test) {
         BOOST_REQUIRE(is_partition_dead(alpha));
     });
 }
+
+SEASTAR_TEST_CASE(sstable_needs_cleanup_test) {
+    test_env env;
+    auto s = make_lw_shared(schema({}, some_keyspace, some_column_family,
+        {{"p1", utf8_type}}, {}, {}, {}, utf8_type));
+
+    auto tokens = token_generation_for_current_shard(10);
+
+    auto sst_gen = [&env, s, gen = make_lw_shared<unsigned>(1)] (sstring first, sstring last) mutable {
+        return sstable_for_overlapping_test(env, s, (*gen)++, first, last);
+    };
+    auto token = [&] (size_t index) -> dht::token {
+        return tokens[index].second;
+    };
+    auto key_from_token = [&] (size_t index) -> sstring {
+        return tokens[index].first;
+    };
+    auto token_range = [&] (size_t first, size_t last) -> dht::token_range {
+        return dht::token_range::make(token(first), token(last));
+    };
+
+    {
+        auto local_ranges = { token_range(0, 9) };
+        auto sst = sst_gen(key_from_token(0), key_from_token(9));
+        BOOST_REQUIRE(!needs_cleanup(sst, local_ranges, s));
+    }
+
+    {
+        auto local_ranges = { token_range(0, 1), token_range(3, 4), token_range(5, 6) };
+
+        auto sst = sst_gen(key_from_token(0), key_from_token(1));
+        BOOST_REQUIRE(!needs_cleanup(sst, local_ranges, s));
+
+        auto sst2 = sst_gen(key_from_token(2), key_from_token(2));
+        BOOST_REQUIRE(needs_cleanup(sst2, local_ranges, s));
+
+        auto sst3 = sst_gen(key_from_token(0), key_from_token(6));
+        BOOST_REQUIRE(needs_cleanup(sst3, local_ranges, s));
+
+        auto sst5 = sst_gen(key_from_token(7), key_from_token(7));
+        BOOST_REQUIRE(needs_cleanup(sst5, local_ranges, s));
+    }
+
+    return make_ready_future<>();
+}

test/boost/stall_free_test.cc

@@ -0,0 +1,55 @@
+/*
+ * Copyright (C) 2020 ScyllaDB
+ */
+
+/*
+ * This file is part of Scylla.
+ *
+ * Scylla is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Scylla is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <seastar/testing/thread_test_case.hh>
+#include "utils/stall_free.hh"
+
+SEASTAR_THREAD_TEST_CASE(test_merge1) {
+    std::list<int> l1{1, 2, 5, 8};
+    std::list<int> l2{3};
+    std::list<int> expected{1,2,3,5,8};
+    utils::merge_to_gently(l1, l2, std::less<int>());
+    BOOST_CHECK(l1 == expected);
+}
+
+SEASTAR_THREAD_TEST_CASE(test_merge2) {
+    std::list<int> l1{1};
+    std::list<int> l2{3, 5, 6};
+    std::list<int> expected{1,3,5,6};
+    utils::merge_to_gently(l1, l2, std::less<int>());
+    BOOST_CHECK(l1 == expected);
+}
+
+SEASTAR_THREAD_TEST_CASE(test_merge3) {
+    std::list<int> l1{};
+    std::list<int> l2{3, 5, 6};
+    std::list<int> expected{3,5,6};
+    utils::merge_to_gently(l1, l2, std::less<int>());
+    BOOST_CHECK(l1 == expected);
+}
+
+SEASTAR_THREAD_TEST_CASE(test_merge4) {
+    std::list<int> l1{1};
+    std::list<int> l2{};
+    std::list<int> expected{1};
+    utils::merge_to_gently(l1, l2, std::less<int>());
+    BOOST_CHECK(l1 == expected);
+}

test/boost/view_schema_test.cc

@@ -118,6 +118,25 @@ SEASTAR_TEST_CASE(test_access_and_schema) {
     });
 }
 
+SEASTAR_TEST_CASE(test_column_dropped_from_base) {
+    return do_with_cql_env_thread([] (auto& e) {
+        e.execute_cql("create table cf (p int, c ascii, a int, v int, primary key (p, c));").get();
+        e.execute_cql("create materialized view vcf as select p, c, v from cf "
+                      "where v is not null and p is not null and c is not null "
+                      "primary key (v, p, c)").get();
+        e.execute_cql("alter table cf drop a;").get();
+        e.execute_cql("insert into cf (p, c, v) values (0, 'foo', 1);").get();
+        eventually([&] {
+        auto msg = e.execute_cql("select v from vcf").get0();
+        assert_that(msg).is_rows()
+            .with_size(1)
+            .with_row({
+                {int32_type->decompose(1)}
+            });
+        });
+    });
+}
+
 SEASTAR_TEST_CASE(test_updates) {
     return do_with_cql_env_thread([] (auto& e) {
         e.execute_cql("create table base (k int, v int, primary key (k));").get();

test/cql/counters_test.cql

@@ -0,0 +1,29 @@
+CREATE TABLE ks.tbl_cnt (pk int PRIMARY KEY, c1 counter, c2 counter);
+
+-- insert some values in one column
+UPDATE ks.tbl_cnt SET c1 = c1+1 WHERE pk = 1;
+UPDATE ks.tbl_cnt SET c1 = c1+2 WHERE pk = 2;
+UPDATE ks.tbl_cnt SET c1 = c1+3 WHERE pk = 3;
+UPDATE ks.tbl_cnt SET c1 = c1+4 WHERE pk = 4;
+
+-- test various filtering options on counter column
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 < 3 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 < 1 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 <= 3 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > 2 AND pk = 4 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 >= 3 and pk = 3 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > 4 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 in (-1, 2, 3) ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 0 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 1 ALLOW FILTERING;
+
+-- now filter through untouched counters `c2` - they should appear as NULLs and evaluate as zeros
+SELECT pk, c1, c2 FROM ks.tbl_cnt WHERE c2 = 0 ALLOW FILTERING;
+SELECT pk, c2 FROM ks.tbl_cnt WHERE c2 < 0 ALLOW FILTERING;
+SELECT pk, c2 FROM ks.tbl_cnt WHERE c2 > 0 ALLOW FILTERING;
+
+-- delete `c1` and make sure it doesn't appear in filtering results
+DELETE c1 from ks.tbl_cnt WHERE pk = 1;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 1 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 <= 1000 ALLOW FILTERING;
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > -1000 ALLOW FILTERING;

test/cql/counters_test.result

@@ -0,0 +1,190 @@
+CREATE TABLE ks.tbl_cnt (pk int PRIMARY KEY, c1 counter, c2 counter);
+{
+	"status" : "ok"
+}
+
+-- insert some values in one column
+UPDATE ks.tbl_cnt SET c1 = c1+1 WHERE pk = 1;
+{
+	"status" : "ok"
+}
+UPDATE ks.tbl_cnt SET c1 = c1+2 WHERE pk = 2;
+{
+	"status" : "ok"
+}
+UPDATE ks.tbl_cnt SET c1 = c1+3 WHERE pk = 3;
+{
+	"status" : "ok"
+}
+UPDATE ks.tbl_cnt SET c1 = c1+4 WHERE pk = 4;
+{
+	"status" : "ok"
+}
+
+-- test various filtering options on counter column
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 < 3 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "1",
+			"pk" : "1"
+		},
+		{
+			"c1" : "2",
+			"pk" : "2"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 < 1 ALLOW FILTERING;
+{
+	"rows" : null
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 <= 3 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "1",
+			"pk" : "1"
+		},
+		{
+			"c1" : "2",
+			"pk" : "2"
+		},
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > 2 AND pk = 4 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "4",
+			"pk" : "4"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 >= 3 and pk = 3 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > 4 ALLOW FILTERING;
+{
+	"rows" : null
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 in (-1, 2, 3) ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "2",
+			"pk" : "2"
+		},
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 0 ALLOW FILTERING;
+{
+	"rows" : null
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 1 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "1",
+			"pk" : "1"
+		}
+	]
+}
+
+-- now filter through untouched counters `c2` - they should appear as NULLs and evaluate as zeros
+SELECT pk, c1, c2 FROM ks.tbl_cnt WHERE c2 = 0 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "1",
+			"pk" : "1"
+		},
+		{
+			"c1" : "2",
+			"pk" : "2"
+		},
+		{
+			"c1" : "4",
+			"pk" : "4"
+		},
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}
+SELECT pk, c2 FROM ks.tbl_cnt WHERE c2 < 0 ALLOW FILTERING;
+{
+	"rows" : null
+}
+SELECT pk, c2 FROM ks.tbl_cnt WHERE c2 > 0 ALLOW FILTERING;
+{
+	"rows" : null
+}
+
+-- delete `c1` and make sure it doesn't appear in filtering results
+DELETE c1 from ks.tbl_cnt WHERE pk = 1;
+{
+	"status" : "ok"
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 = 1 ALLOW FILTERING;
+{
+	"rows" : null
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 <= 1000 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "2",
+			"pk" : "2"
+		},
+		{
+			"c1" : "4",
+			"pk" : "4"
+		},
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}
+SELECT pk, c1 FROM ks.tbl_cnt WHERE c1 > -1000 ALLOW FILTERING;
+{
+	"rows" : 
+	[
+		{
+			"c1" : "2",
+			"pk" : "2"
+		},
+		{
+			"c1" : "4",
+			"pk" : "4"
+		},
+		{
+			"c1" : "3",
+			"pk" : "3"
+		}
+	]
+}

test/lib/sstable_utils.cc

@@ -121,6 +121,14 @@ token_generation_for_shard(unsigned tokens_to_generate, unsigned shard,
     return key_and_token_pair;
 }
 
+future<compaction_info> compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf, std::function<shared_sstable()> creator, replacer_fn replacer) {
+    descriptor.creator = [creator = std::move(creator)] (shard_id dummy) mutable {
+        return creator();
+    };
+    descriptor.replacer = std::move(replacer);
+    return sstables::compact_sstables(std::move(descriptor), cf);
+}
+
 std::vector<std::pair<sstring, dht::token>> token_generation_for_current_shard(unsigned tokens_to_generate) {
     return token_generation_for_shard(tokens_to_generate, engine().cpu_id());
 }

test/lib/sstable_utils.hh

@@ -355,5 +355,7 @@ public:
     }
 };
 
-
 } // namespace sstables
+
+future<compaction_info> compact_sstables(sstables::compaction_descriptor descriptor, column_family& cf,
+        std::function<shared_sstable()> creator, replacer_fn replacer = sstables::replacer_fn_no_op());